sources.jsonl

{"id": "glm-4.5-paper", "type": "paper", "title": "GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models", "url": "https://arxiv.org/abs/2508.06471", "arxiv_id": "2508.06471", "models": ["zai-org/GLM-4.5", "zai-org/GLM-4.5-Air", "zai-org/GLM-4.6", "zai-org/GLM-4.7", "zai-org/GLM-4.7-Flash"], "notes": "Paper describes GLM-4.5; later GLM-4.x releases reuse the same tag/suite on the Hub.", "benchmarks": [{"name": "SimpleQA", "category": "knowledge", "models": null}, {"name": "MMLU", "category": "knowledge", "models": null}, {"name": "TriviaQA", "category": "knowledge", "models": null}, {"name": "Chinese-SimpleQA", "category": "knowledge", "models": null}, {"name": "C-Eval", "category": "knowledge", "models": null}, {"name": "C3", "category": "knowledge", "models": null}, {"name": "BBH", "category": "reasoning", "models": null}, {"name": "HellaSwag", "category": "reasoning", "models": null}, {"name": "PIQA", "category": "reasoning", "models": null}, {"name": "MMLU-Pro", "category": "reasoning", "models": null}, {"name": "GPQA", "category": "reasoning", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "CLUEWSC", "category": "reasoning", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "AIME 2024", "category": "math", "models": null}, {"name": "MATH-500", "category": "math", "models": null}, {"name": "EvalPlus", "category": "coding", "models": null}, {"name": "LiveCodeBench-Base", "category": "coding", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "Terminal-Bench", "category": "coding", "models": null}, {"name": "SciCode", "category": "coding", "models": null}, {"name": "TAU-Bench", "category": "agentic", "models": null}, {"name": "BFCL v3", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "SysBench", "category": "general", "models": null}, {"name": "MultiChallenge", "category": "general", "models": null}, {"name": "SafetyBench", "category": "safety", "models": null}, {"name": "CC-Bench", "category": "human_eval", "models": null}]}
{"id": "glm-5-paper", "type": "paper", "title": "GLM-5: from Vibe Coding to Agentic Engineering", "url": "https://arxiv.org/abs/2602.15763", "arxiv_id": "2602.15763", "models": ["zai-org/GLM-5", "zai-org/GLM-5.1", "zai-org/GLM-5.2"], "notes": null, "benchmarks": [{"name": "HLE", "category": "reasoning", "models": null}, {"name": "HLE w/ Tools", "category": "reasoning", "models": null}, {"name": "AIME 2026", "category": "reasoning", "models": null}, {"name": "HMMT Feb 2025", "category": "reasoning", "models": null}, {"name": "HMMT Nov 2025", "category": "reasoning", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "LongBench v2", "category": "reasoning", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "CyberGym", "category": "coding", "models": null}, {"name": "SWE-rebench", "category": "coding", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "BrowseComp-ZH", "category": "agentic", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "MCP-Atlas", "category": "agentic", "models": null}, {"name": "Tool-Decathlon", "category": "agentic", "models": null}, {"name": "Vending-Bench 2", "category": "agentic", "models": null}, {"name": "GDPval-AA", "category": "agentic", "models": null}, {"name": "CC-Bench-V2", "category": "agentic", "models": null}, {"name": "SimpleQA", "category": "knowledge", "models": null}, {"name": "BBH", "category": "knowledge", "models": null}, {"name": "MMLU", "category": "knowledge", "models": null}, {"name": "HellaSwag", "category": "knowledge", "models": null}, {"name": "PIQA", "category": "knowledge", "models": null}, {"name": "TriviaQA", "category": "knowledge", "models": null}, {"name": "C-Eval", "category": "knowledge", "models": null}, {"name": "C3", "category": "knowledge", "models": null}, {"name": "C-SimpleQA", "category": "knowledge", "models": null}, {"name": "EvalPlus", "category": "coding_base", "models": null}, {"name": "LiveCodeBench-Base", "category": "coding_base", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "ZMultiTransBench", "category": "multilingual", "models": null}, {"name": "LMArena", "category": "general", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "MultiChallenge", "category": "general", "models": null}, {"name": "ToolCall-Badcase", "category": "general", "models": null}]}
{"id": "kimi-k2.5-paper", "type": "paper", "title": "Kimi K2.5: Visual Agentic Intelligence", "url": "https://arxiv.org/abs/2602.02276", "arxiv_id": "2602.02276", "models": ["moonshotai/Kimi-K2.5", "moonshotai/Kimi-K2.6"], "notes": null, "benchmarks": [{"name": "HLE", "category": "reasoning", "models": null}, {"name": "AIME 2025", "category": "reasoning", "models": null}, {"name": "HMMT Feb 2025", "category": "reasoning", "models": null}, {"name": "HMMT Nov 2025", "category": "reasoning", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "MMLU-Pro", "category": "reasoning", "models": null}, {"name": "SimpleQA Verified", "category": "reasoning", "models": null}, {"name": "AdvancedIF", "category": "reasoning", "models": null}, {"name": "LongBench v2", "category": "reasoning", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Pro", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "PaperBench", "category": "coding", "models": null}, {"name": "CyberGym", "category": "coding", "models": null}, {"name": "SciCode", "category": "coding", "models": null}, {"name": "OJBench", "category": "coding", "models": null}, {"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "WideSearch", "category": "agentic", "models": null}, {"name": "DeepSearchQA", "category": "agentic", "models": null}, {"name": "FinSearchComp", "category": "agentic", "models": null}, {"name": "Seal-0", "category": "agentic", "models": null}, {"name": "GDPval", "category": "agentic", "models": null}, {"name": "MMMU", "category": "vision", "models": null}, {"name": "MMMU-Pro", "category": "vision", "models": null}, {"name": "CharXiv", "category": "vision", "models": null}, {"name": "MathVision", "category": "vision", "models": null}, {"name": "MathVista", "category": "vision", "models": null}, {"name": "ZeroBench", "category": "vision", "models": null}, {"name": "BLINK", "category": "vision", "models": null}, {"name": "OCRBench", "category": "vision", "models": null}, {"name": "OmniDocBench", "category": "vision", "models": null}, {"name": "InfoVQA", "category": "vision", "models": null}, {"name": "VideoMMMU", "category": "video", "models": null}, {"name": "MMVU", "category": "video", "models": null}, {"name": "MotionBench", "category": "video", "models": null}, {"name": "Video-MME", "category": "video", "models": null}, {"name": "LongVideoBench", "category": "video", "models": null}, {"name": "LVBench", "category": "video", "models": null}, {"name": "OSWorld-Verified", "category": "computer_use", "models": null}, {"name": "WebArena", "category": "computer_use", "models": null}]}
{"id": "deepseek-v4-paper", "type": "paper", "title": "DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence", "url": "https://arxiv.org/abs/2606.19348", "arxiv_id": "2606.19348", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash", "deepseek-ai/DeepSeek-V4-Flash-0731"], "notes": "DeepSeek-V4-Flash-0731 is the official release superseding the V4-Flash preview (same arXiv tag); its agentic benchmark table comes from its own HF model card and is scoped to it ('Agents' Last Exam' -> Agent's Last Exam; 'AutomationBench Public' -> AutomationBench, same 600-task public subset other sources use). DSBench-FullStack and DSBench-Hard are internal DeepSeek test sets. The paper suite is scoped to the V4-Pro/V4-Flash preview models.", "benchmarks": [{"name": "AGIEval", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "C-Eval", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "CMMLU", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MMLU", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MMLU-Redux", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MMLU-Pro", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MMMLU", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "SimpleQA", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "SimpleQA Verified", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "C-SimpleQA", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "SuperGPQA", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "GPQA-Diamond", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "HLE", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "TriviaQA", "category": "knowledge", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "BBH", "category": "reasoning", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "DROP", "category": "reasoning", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "HellaSwag", "category": "reasoning", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "WinoGrande", "category": "reasoning", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "BigCodeBench", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "HumanEval", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "LiveCodeBench v6", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "Codeforces", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "SWE-bench Verified", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "SWE-bench Multilingual", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "SWE-bench Pro", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "GSM8K", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MATH", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MGSM", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "HMMT", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "Apex", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "IMO-AnswerBench", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "PutnamBench", "category": "math", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "LongBench v2", "category": "long_context", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MRCR", "category": "long_context", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "CorpusQA", "category": "long_context", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "BrowseComp", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "MCP-Atlas", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "GDPval-AA", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "Tool-Decathlon", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Pro", "deepseek-ai/DeepSeek-V4-Flash"]}, {"name": "Terminal-Bench 2.1", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "NL2Repo", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "CyberGym", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "DeepSWE", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "DSBench-FullStack", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "DSBench-Hard", "category": "coding", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "Toolathlon-Verified", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "Agent's Last Exam", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}, {"name": "AutomationBench", "category": "agentic", "models": ["deepseek-ai/DeepSeek-V4-Flash-0731"]}]}
{"id": "qwen3.5-flagship-card", "type": "model_card", "title": "Qwen3.5-397B-A17B / Qwen3.5-122B-A10B model card", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "arxiv_id": null, "models": ["Qwen/Qwen3.5-397B-A17B", "Qwen/Qwen3.5-122B-A10B"], "notes": null, "benchmarks": [{"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "MMLU-Redux", "category": "knowledge", "models": null}, {"name": "SuperGPQA", "category": "knowledge", "models": null}, {"name": "C-Eval", "category": "knowledge", "models": null}, {"name": "MMMLU", "category": "knowledge", "models": null}, {"name": "MMLU-ProX", "category": "knowledge", "models": null}, {"name": "NOVA-63", "category": "knowledge", "models": null}, {"name": "INCLUDE", "category": "knowledge", "models": null}, {"name": "Global PIQA", "category": "knowledge", "models": null}, {"name": "PolyMATH", "category": "knowledge", "models": null}, {"name": "WMT24++", "category": "knowledge", "models": null}, {"name": "MAXIFE", "category": "knowledge", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "MultiChallenge", "category": "general", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "LongBench v2", "category": "long_context", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "HLE Verified", "category": "reasoning", "models": null}, {"name": "LiveCodeBench v6", "category": "reasoning", "models": null}, {"name": "HMMT Feb 2025", "category": "reasoning", "models": null}, {"name": "HMMT Nov 2025", "category": "reasoning", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning", "models": null}, {"name": "AIME 2026", "category": "reasoning", "models": null}, {"name": "BFCL-V4", "category": "agentic", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "VITA-Bench", "category": "agentic", "models": null}, {"name": "DeepPlanning", "category": "agentic", "models": null}, {"name": "Tool-Decathlon", "category": "agentic", "models": null}, {"name": "MCP-Mark", "category": "agentic", "models": null}, {"name": "HLE w/ Tools", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "BrowseComp-ZH", "category": "agentic", "models": null}, {"name": "WideSearch", "category": "agentic", "models": null}, {"name": "Seal-0", "category": "agentic", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "SecCodeBench", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "Codeforces", "category": "coding", "models": null}, {"name": "OJBench", "category": "coding", "models": null}, {"name": "FullStackBench", "category": "coding", "models": null}, {"name": "MMMU", "category": "vision_language", "models": null}, {"name": "MMMU-Pro", "category": "vision_language", "models": null}, {"name": "MathVision", "category": "vision_language", "models": null}, {"name": "MathVista", "category": "vision_language", "models": null}, {"name": "ZeroBench", "category": "vision_language", "models": null}, {"name": "DynaMath", "category": "vision_language", "models": null}, {"name": "We-Math", "category": "vision_language", "models": null}, {"name": "MedXpertQA-MM", "category": "vision_language", "models": null}, {"name": "CharXiv", "category": "vision_language", "models": null}, {"name": "MMStar", "category": "vision_language", "models": null}, {"name": "RealWorldQA", "category": "vision_language", "models": null}, {"name": "HallusionBench", "category": "vision_language", "models": null}, {"name": "OCRBench", "category": "vision_language", "models": null}, {"name": "CC-OCR", "category": "vision_language", "models": null}, {"name": "OmniDocBench", "category": "vision_language", "models": null}, {"name": "MMLongBench-Doc", "category": "vision_language", "models": null}, {"name": "AI2D", "category": "vision_language", "models": null}, {"name": "PMC-VQA", "category": "vision_language", "models": null}, {"name": "SLAKE", "category": "vision_language", "models": null}, {"name": "SimpleVQA", "category": "vision_language", "models": null}, {"name": "MLVU", "category": "vision_language", "models": null}, {"name": "MVBench", "category": "vision_language", "models": null}, {"name": "LVBench", "category": "vision_language", "models": null}, {"name": "VideoMMMU", "category": "vision_language", "models": null}, {"name": "Video-MME", "category": "vision_language", "models": null}, {"name": "LingoQA", "category": "vision_language", "models": null}, {"name": "MMVU", "category": "vision_language", "models": null}, {"name": "ERQA", "category": "vision_language", "models": null}, {"name": "EmbSpatialBench", "category": "vision_language", "models": null}, {"name": "RefCOCO", "category": "vision_language", "models": null}, {"name": "RefSpatialBench", "category": "vision_language", "models": null}, {"name": "ScreenSpot Pro", "category": "vision_language", "models": null}, {"name": "AndroidWorld", "category": "vision_language", "models": null}, {"name": "OSWorld-Verified", "category": "vision_language", "models": null}, {"name": "Hypersim", "category": "vision_language", "models": null}, {"name": "SUNRGBD", "category": "vision_language", "models": null}, {"name": "Nuscenes", "category": "vision_language", "models": null}, {"name": "ODinW13", "category": "vision_language", "models": null}, {"name": "CountBench", "category": "vision_language", "models": null}, {"name": "BabyVision", "category": "vision_language", "models": null}, {"name": "V*", "category": "vision_language", "models": null}, {"name": "TIR-Bench", "category": "vision_language", "models": null}]}
{"id": "qwen3.6-27b-card", "type": "model_card", "title": "Qwen3.6-27B model card", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "arxiv_id": null, "models": ["Qwen/Qwen3.6-27B"], "notes": null, "benchmarks": [{"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Pro", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "SkillsBench", "category": "coding", "models": null}, {"name": "QwenWebBench", "category": "coding", "models": null}, {"name": "NL2Repo", "category": "coding", "models": null}, {"name": "Claw-Eval", "category": "coding", "models": null}, {"name": "QwenClawBench", "category": "coding", "models": null}, {"name": "AndroidWorld", "category": "coding", "models": null}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "MMLU-Redux", "category": "knowledge", "models": null}, {"name": "SuperGPQA", "category": "knowledge", "models": null}, {"name": "C-Eval", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "LiveCodeBench v6", "category": "reasoning", "models": null}, {"name": "HMMT Feb 2025", "category": "reasoning", "models": null}, {"name": "HMMT Nov 2025", "category": "reasoning", "models": null}, {"name": "HMMT Feb 2026", "category": "reasoning", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning", "models": null}, {"name": "AIME 2026", "category": "reasoning", "models": null}, {"name": "MMMU", "category": "vision_language", "models": null}, {"name": "MMMU-Pro", "category": "vision_language", "models": null}, {"name": "MathVista", "category": "vision_language", "models": null}, {"name": "DynaMath", "category": "vision_language", "models": null}, {"name": "VlmsAreBlind", "category": "vision_language", "models": null}, {"name": "RealWorldQA", "category": "vision_language", "models": null}, {"name": "MMStar", "category": "vision_language", "models": null}, {"name": "MMBench", "category": "vision_language", "models": null}, {"name": "SimpleVQA", "category": "vision_language", "models": null}, {"name": "CharXiv", "category": "vision_language", "models": null}, {"name": "CC-OCR", "category": "vision_language", "models": null}, {"name": "OCRBench", "category": "vision_language", "models": null}, {"name": "ERQA", "category": "vision_language", "models": null}, {"name": "CountBench", "category": "vision_language", "models": null}, {"name": "RefCOCO", "category": "vision_language", "models": null}, {"name": "EmbSpatialBench", "category": "vision_language", "models": null}, {"name": "RefSpatialBench", "category": "vision_language", "models": null}, {"name": "Video-MME", "category": "vision_language", "models": null}, {"name": "VideoMMMU", "category": "vision_language", "models": null}, {"name": "MLVU", "category": "vision_language", "models": null}, {"name": "MVBench", "category": "vision_language", "models": null}, {"name": "V*", "category": "vision_language", "models": null}]}
{"id": "nemotron-3-ultra-report", "type": "report", "title": "NVIDIA Nemotron 3 Ultra Technical Report", "url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf", "arxiv_id": null, "models": ["nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4"], "notes": null, "benchmarks": [{"name": "Terminal-Bench 2.1", "category": "agentic", "models": null}, {"name": "GDPval", "category": "agentic", "models": null}, {"name": "SWE-bench Verified", "category": "agentic", "models": null}, {"name": "SWE-bench Multilingual", "category": "agentic", "models": null}, {"name": "ProfBench", "category": "agentic", "models": null}, {"name": "PinchBench", "category": "agentic", "models": null}, {"name": "TAU3-Bench", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "AGIEval", "category": "knowledge", "models": null}, {"name": "GPQA", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "AA-Omniscience", "category": "knowledge", "models": null}, {"name": "SciCode", "category": "knowledge", "models": null}, {"name": "CritPt", "category": "knowledge", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "IMO-AnswerBench", "category": "math", "models": null}, {"name": "MathArena Apex", "category": "math", "models": null}, {"name": "Putnam 2025", "category": "math", "models": null}, {"name": "USAMO 2026", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "IOI 2025", "category": "coding", "models": null}, {"name": "ARC-Challenge", "category": "commonsense", "models": null}, {"name": "OpenBookQA", "category": "commonsense", "models": null}, {"name": "PIQA", "category": "commonsense", "models": null}, {"name": "HellaSwag", "category": "commonsense", "models": null}, {"name": "WinoGrande", "category": "commonsense", "models": null}, {"name": "RULER", "category": "long_context", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "LongBench v2", "category": "long_context", "models": null}, {"name": "MGSM", "category": "multilingual", "models": null}, {"name": "Global-MMLU-Lite", "category": "multilingual", "models": null}, {"name": "MMLU-ProX", "category": "multilingual", "models": null}, {"name": "WMT24++", "category": "multilingual", "models": null}]}
{"id": "longcat-2.0-blog", "type": "blog", "title": "LongCat-2.0 announcement blog / HF discussion", "url": "https://longcat.chat/blog/longcat-2.0", "arxiv_id": null, "models": ["meituan-longcat/LongCat-2.0"], "notes": "HF model card has no benchmark table; numbers are blog-only self-reported. Only SWE-bench Pro and GPQA-Diamond have Hub-registered eval.yaml mappings.", "benchmarks": [{"name": "SWE-bench Pro", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "RWSearch", "category": "agentic", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "FORTE", "category": "reasoning", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "Writing Bench", "category": "general", "models": null}]}
{"id": "hy3-preview-card", "type": "model_card", "title": "Hy3-preview GitHub README / model card", "url": "https://github.com/Tencent-Hunyuan/Hy3-preview", "arxiv_id": null, "models": ["tencent/Hy3-preview"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "pretrained", "models": null}, {"name": "MMLU-Pro", "category": "pretrained", "models": null}, {"name": "MMLU-Redux", "category": "pretrained", "models": null}, {"name": "ARC-Challenge", "category": "pretrained", "models": null}, {"name": "DROP", "category": "pretrained", "models": null}, {"name": "PIQA", "category": "pretrained", "models": null}, {"name": "SuperGPQA", "category": "pretrained", "models": null}, {"name": "SimpleQA", "category": "pretrained", "models": null}, {"name": "GSM8K", "category": "pretrained", "models": null}, {"name": "MATH", "category": "pretrained", "models": null}, {"name": "CMath", "category": "pretrained", "models": null}, {"name": "LiveCodeBench v6", "category": "pretrained", "models": null}, {"name": "FrontierScience-Olympiad", "category": "reasoning", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "Tsinghua Qiuzhen Math PhD Exam", "category": "reasoning", "models": null}, {"name": "CHSBO 2025", "category": "reasoning", "models": null}, {"name": "AdvancedIF", "category": "general", "models": null}, {"name": "AA-LCR", "category": "general", "models": null}, {"name": "LongBench v2", "category": "general", "models": null}, {"name": "SWE-bench Verified", "category": "agentic", "models": null}, {"name": "Terminal-Bench 2.0", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "WideSearch", "category": "agentic", "models": null}, {"name": "WildClawBench", "category": "agentic", "models": null}, {"name": "Claw-Eval", "category": "agentic", "models": null}]}
{"id": "internscience-agents-a1-paper", "type": "paper", "title": "Agents-A1 technical report", "url": "https://arxiv.org/abs/2606.30616", "arxiv_id": "2606.30616", "models": ["InternScience/Agents-A1"], "notes": null, "benchmarks": [{"name": "GAIA", "category": "agentic_search", "models": null}, {"name": "BrowseComp", "category": "agentic_search", "models": null}, {"name": "xBench-DeepSearch", "category": "agentic_search", "models": null}, {"name": "Seal-0", "category": "agentic_search", "models": null}, {"name": "SciCode", "category": "engineering", "models": null}, {"name": "MLE-Bench-Lite", "category": "engineering", "models": null}, {"name": "HLE", "category": "scientific", "models": null}, {"name": "HiPhO", "category": "scientific", "models": null}, {"name": "FrontierScience-Olympiad", "category": "scientific", "models": null}, {"name": "FrontierScience-Research", "category": "scientific", "models": null}, {"name": "LongBench v2", "category": "long_context", "models": null}, {"name": "IFBench", "category": "long_context", "models": null}, {"name": "IFEval", "category": "long_context", "models": null}, {"name": "TAU2-Bench", "category": "agentic_tool_use", "models": null}, {"name": "VITA-Bench", "category": "agentic_tool_use", "models": null}, {"name": "MolBench", "category": "domain_specific", "models": null}, {"name": "MatTools", "category": "domain_specific", "models": null}]}
{"id": "k-exaone-paper", "type": "paper", "title": "K-EXAONE-236B-A23B technical report", "url": "https://arxiv.org/abs/2601.01739", "arxiv_id": "2601.01739", "models": ["LGAI-EXAONE/K-EXAONE-236B-A23B"], "notes": null, "benchmarks": [{"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "IMO-AnswerBench", "category": "math", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "HMMT Nov 2025", "category": "math", "models": null}, {"name": "LiveCodeBench Pro", "category": "coding", "models": null}, {"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "CodeUtilityBench", "category": "coding", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "MRCR", "category": "long_context", "models": null}, {"name": "Needle-in-a-haystack", "category": "long_context", "models": null}, {"name": "KMMLU-Pro", "category": "korean", "models": null}, {"name": "KoBALT", "category": "korean", "models": null}, {"name": "CLIcK", "category": "korean", "models": null}, {"name": "HRM8K", "category": "korean", "models": null}, {"name": "Ko-LongBench", "category": "korean", "models": null}, {"name": "MMMLU", "category": "multilingual", "models": null}, {"name": "WMT24++", "category": "multilingual", "models": null}, {"name": "WildJailbreak", "category": "safety", "models": null}, {"name": "KGC-Safety", "category": "safety", "models": null}]}
{"id": "lfm2-technical-report", "type": "paper", "title": "LFM2 Technical Report", "url": "https://arxiv.org/abs/2511.23404", "arxiv_id": "2511.23404", "models": ["LiquidAI/LFM2.5-350M", "LiquidAI/LFM2.5-1.2B-Instruct"], "notes": "LFM2.5-1.2B-Instruct benchmarks (GPQA, MMLU-Pro, IFEval, IFBench, Multi-IF, AIME 2025, BFCL v3) come from its own HF model card, which cites the same LFM2 Technical Report (arXiv 2511.23404); card's 'AIME25'/'BFCLv3' canonicalized. The math/multilingual/MMLU suite is scoped to LFM2.5-350M (from the report).", "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "GPQA", "category": "knowledge", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "Multi-IF", "category": "general", "models": null}, {"name": "GSM8K", "category": "math", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "GSMPlus", "category": "math", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "MATH-500", "category": "math", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "MATH Level 5", "category": "math", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "MMMLU", "category": "multilingual", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "MGSM", "category": "multilingual", "models": ["LiquidAI/LFM2.5-350M"]}, {"name": "AIME 2025", "category": "math", "models": ["LiquidAI/LFM2.5-1.2B-Instruct"]}, {"name": "BFCL v3", "category": "agentic", "models": ["LiquidAI/LFM2.5-1.2B-Instruct"]}]}
{"id": "nanbeige4.1-paper", "type": "paper", "title": "Nanbeige4.1-3B technical report", "url": "https://arxiv.org/abs/2602.13367", "arxiv_id": "2602.13367", "models": ["Nanbeige/Nanbeige4.1-3B"], "notes": null, "benchmarks": [{"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "LCB-Pro-Easy", "category": "coding", "models": null}, {"name": "LCB-Pro-Medium", "category": "coding", "models": null}, {"name": "AIME 2026-I", "category": "math", "models": null}, {"name": "HMMT Nov", "category": "math", "models": null}, {"name": "IMO-AnswerBench", "category": "math", "models": null}, {"name": "GPQA", "category": "science", "models": null}, {"name": "HLE", "category": "science", "models": null}, {"name": "Arena-Hard-V2", "category": "alignment", "models": null}, {"name": "MultiChallenge", "category": "alignment", "models": null}, {"name": "BFCL-V4", "category": "agentic", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "GAIA", "category": "deep_search", "models": null}, {"name": "BrowseComp", "category": "deep_search", "models": null}, {"name": "BrowseComp-ZH", "category": "deep_search", "models": null}, {"name": "Seal-0", "category": "deep_search", "models": null}, {"name": "xBench-DeepSearch", "category": "deep_search", "models": null}, {"name": "LeetCode Weekly Contests", "category": "coding_stress_test", "models": null}]}
{"id": "openseeker-v2-paper", "type": "paper", "title": "OpenSeeker-v2-30B-SFT technical report", "url": "https://arxiv.org/abs/2605.04036", "arxiv_id": "2605.04036", "models": ["PolarSeeker/OpenSeeker-v2-30B-SFT"], "notes": null, "benchmarks": [{"name": "BrowseComp", "category": "deep_search_agentic", "models": null}, {"name": "BrowseComp-ZH", "category": "deep_search_agentic", "models": null}, {"name": "HLE", "category": "deep_search_agentic", "models": null}, {"name": "xBench-DeepSearch", "category": "deep_search_agentic", "models": null}]}
{"id": "ax-k1-paper", "type": "paper", "title": "A.X-K1 technical report", "url": "https://arxiv.org/abs/2601.09200", "arxiv_id": "2601.09200", "models": ["skt/A.X-K1"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "KMMLU", "category": "knowledge", "models": null}, {"name": "KMMLU-Redux", "category": "knowledge", "models": null}, {"name": "KMMLU-Pro", "category": "knowledge", "models": null}, {"name": "CLIcK", "category": "knowledge", "models": null}, {"name": "KoBALT", "category": "knowledge", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "AIME25-ko", "category": "math", "models": null}, {"name": "HRM8K", "category": "math", "models": null}, {"name": "HumanEval+", "category": "coding", "models": null}, {"name": "HumanEval+ ko", "category": "coding", "models": null}, {"name": "MBPP+", "category": "coding", "models": null}, {"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "LiveCodeBench-ko", "category": "coding", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "IFEval-ko", "category": "general", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}]}
{"id": "minimax-m2-family", "type": "model_card", "title": "MiniMax-M2 / M2.1 model card (tagged scaffold papers: R2E-Gym 2504.07164, WebExplorer 2509.06501, FinSearchComp 2509.13160)", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "arxiv_id": null, "models": ["MiniMaxAI/MiniMax-M2", "MiniMaxAI/MiniMax-M2.1"], "notes": "No dedicated M2 technical report; tagged arxiv IDs are scaffold/harness methodology references, not the model's own eval report.", "benchmarks": [{"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "Multi-SWE-bench", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "Terminal-Bench Hard", "category": "coding", "models": null}, {"name": "ArtifactsBench", "category": "coding", "models": null}, {"name": "SWT-bench", "category": "coding", "models": null}, {"name": "SWE-Perf", "category": "coding", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "BrowseComp", "category": "agentic_search", "models": null}, {"name": "BrowseComp-ZH", "category": "agentic_search", "models": null}, {"name": "GAIA", "category": "agentic_search", "models": null}, {"name": "xBench-DeepSearch", "category": "agentic_search", "models": null}, {"name": "TAU2-Bench", "category": "agentic_search", "models": null}, {"name": "AgentCompany", "category": "agentic_search", "models": null}, {"name": "FinSearchComp-global", "category": "agentic_search", "models": null}, {"name": "AIME 2025", "category": "reasoning", "models": null}, {"name": "MMLU-Pro", "category": "reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "SciCode", "category": "reasoning", "models": null}, {"name": "IFBench", "category": "reasoning", "models": null}, {"name": "AA-LCR", "category": "reasoning", "models": null}]}
{"id": "minimax-m3-paper", "type": "paper", "title": "MiniMax Sparse Attention (M3's tagged technical report; primarily an architecture paper)", "url": "https://arxiv.org/abs/2606.13392", "arxiv_id": "2606.13392", "models": ["MiniMaxAI/MiniMax-M3"], "notes": "Benchmarks are from a 109B MSA-vs-GQA ablation model in the paper, not the production M3 release (M3's own card shows only an image with no text table).", "benchmarks": [{"name": "MMLU", "category": "reasoning", "models": null}, {"name": "MMLU-Pro", "category": "reasoning", "models": null}, {"name": "BBH", "category": "reasoning", "models": null}, {"name": "GPQA Hard", "category": "reasoning", "models": null}, {"name": "ARC-Challenge", "category": "reasoning", "models": null}, {"name": "TriviaQA", "category": "reasoning", "models": null}, {"name": "WinoGrande", "category": "reasoning", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MGSM", "category": "math", "models": null}, {"name": "MathVista", "category": "math", "models": null}, {"name": "OlymMATH", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "EvalPlus", "category": "coding", "models": null}, {"name": "BigCodeBench", "category": "coding", "models": null}, {"name": "MultiPL-E", "category": "coding", "models": null}, {"name": "AI2D", "category": "vision", "models": null}, {"name": "ChartQA", "category": "vision", "models": null}, {"name": "MMMU", "category": "vision", "models": null}, {"name": "OCRBench v2", "category": "vision", "models": null}, {"name": "CharXiv", "category": "vision", "models": null}, {"name": "VisualWebBench", "category": "vision", "models": null}, {"name": "CVBench", "category": "vision", "models": null}, {"name": "EgoSchema", "category": "video", "models": null}, {"name": "LongVideoBench", "category": "video", "models": null}, {"name": "MLVU", "category": "video", "models": null}, {"name": "MMVU", "category": "video", "models": null}, {"name": "Video-MME", "category": "video", "models": null}, {"name": "TemporalBench", "category": "video", "models": null}, {"name": "RULER", "category": "long_context", "models": null}, {"name": "HELMET", "category": "long_context", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "TheAgentCompany", "category": "agentic", "models": null}, {"name": "HLE", "category": "agentic", "models": null}, {"name": "SWE-bench", "category": "agentic", "models": null}]}
{"id": "qwen3-original-paper", "type": "paper", "title": "Qwen3 Technical Report", "url": "https://arxiv.org/abs/2505.09388", "arxiv_id": "2505.09388", "models": ["Qwen/Qwen3-235B-A22B", "Qwen/Qwen3-8B", "Qwen/Qwen3-Coder-480B-A35B-Instruct", "Qwen/Qwen3-Coder-Next", "Qwen/Qwen3-235B-A22B-Instruct-2507", "Qwen/Qwen3-Coder-30B-A3B-Instruct", "Qwen/Qwen3-Next-80B-A3B-Thinking"], "notes": "Qwen3-Coder-480B and Qwen3-Coder-Next have no text benchmark table on their own model cards (image-only charts); Qwen3-Coder-Next's chart filename implies SWE-bench Pro beyond the base Qwen3 suite.", "benchmarks": [{"name": "MMLU", "category": "base_general", "models": null}, {"name": "MMLU-Pro", "category": "base_general", "models": null}, {"name": "MMLU-Redux", "category": "base_general", "models": null}, {"name": "BBH", "category": "base_general", "models": null}, {"name": "SuperGPQA", "category": "base_general", "models": null}, {"name": "GPQA", "category": "base_math_stem", "models": null}, {"name": "GSM8K", "category": "base_math_stem", "models": null}, {"name": "MATH", "category": "base_math_stem", "models": null}, {"name": "EvalPlus", "category": "base_coding", "models": null}, {"name": "MultiPL-E", "category": "base_coding", "models": null}, {"name": "CRUXEval", "category": "base_coding", "models": null}, {"name": "MGSM", "category": "base_multilingual", "models": null}, {"name": "MMMLU", "category": "base_multilingual", "models": null}, {"name": "INCLUDE", "category": "base_multilingual", "models": null}, {"name": "MMLU-Redux", "category": "post_training_knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "post_training_knowledge", "models": null}, {"name": "C-Eval", "category": "post_training_knowledge", "models": null}, {"name": "LiveBench", "category": "post_training_knowledge", "models": null}, {"name": "IFEval", "category": "post_training_alignment", "models": null}, {"name": "Arena-Hard", "category": "post_training_alignment", "models": null}, {"name": "AlignBench", "category": "post_training_alignment", "models": null}, {"name": "Writing Bench", "category": "post_training_alignment", "models": null}, {"name": "MATH-500", "category": "post_training_math_reasoning", "models": null}, {"name": "AIME 2024", "category": "post_training_math_reasoning", "models": null}, {"name": "AIME 2025", "category": "post_training_math_reasoning", "models": null}, {"name": "ZebraLogic", "category": "post_training_math_reasoning", "models": null}, {"name": "AutoLogi", "category": "post_training_math_reasoning", "models": null}, {"name": "BFCL v3", "category": "post_training_agentic_coding", "models": null}, {"name": "LiveCodeBench v5", "category": "post_training_agentic_coding", "models": null}, {"name": "Codeforces", "category": "post_training_agentic_coding", "models": null}, {"name": "Multi-IF", "category": "post_training_multilingual", "models": null}, {"name": "PolyMATH", "category": "post_training_multilingual", "models": null}, {"name": "MLogiQA", "category": "post_training_multilingual", "models": null}, {"name": "SWE-bench Pro", "category": "coder_implied", "models": ["Qwen/Qwen3-Coder-Next"]}]}
{"id": "nemotron-3-nano-super", "type": "paper", "title": "Nemotron 3 Nano (2512.20848) + NVIDIA Nemotron 3 family whitepaper (2512.20856)", "url": "https://arxiv.org/abs/2512.20848", "arxiv_id": "2512.20848", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8", "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-MXFP4_MOE-dequant-bf16-vllm", "exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q3_K_M-dequant-bf16-vllm", "exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-dequant-bf16-vllm"], "notes": "2512.20848 is the Nano-specific technical report. 2512.20856 is a family-wide architecture whitepaper; at time of writing only Nano had shipped, so its benchmark tables are internal ablations, not full Super/Ultra evals. RedHatAI and exolabs repos are community requantizations of the NVIDIA originals and inherit the same suite.", "benchmarks": [{"name": "HumanEval", "category": "base", "models": null}, {"name": "HumanEval+", "category": "base", "models": null}, {"name": "MBPP", "category": "base", "models": null}, {"name": "MBPP+", "category": "base", "models": null}, {"name": "MATH-500", "category": "base", "models": null}, {"name": "GSM8K", "category": "base", "models": null}, {"name": "MMLU", "category": "base", "models": null}, {"name": "MMLU-Pro", "category": "base", "models": null}, {"name": "MMLU-Redux", "category": "base", "models": null}, {"name": "ARC-Challenge", "category": "base", "models": null}, {"name": "HellaSwag", "category": "base", "models": null}, {"name": "WinoGrande", "category": "base", "models": null}, {"name": "RACE", "category": "base", "models": null}, {"name": "Global-MMLU-Lite", "category": "base", "models": null}, {"name": "MGSM", "category": "base", "models": null}, {"name": "RULER", "category": "base", "models": null}, {"name": "AIME 2025", "category": "post_trained_reasoning", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "GPQA", "category": "post_trained_reasoning", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "LiveCodeBench v6", "category": "post_trained_reasoning", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "SciCode", "category": "post_trained_reasoning", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "HLE", "category": "post_trained_reasoning", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "MMLU-Pro", "category": "post_trained_reasoning", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "Terminal-Bench", "category": "post_trained_agentic", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "SWE-bench", "category": "post_trained_agentic", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "TAU2-Bench", "category": "post_trained_agentic", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "BFCL-V4", "category": "post_trained_agentic", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "IFBench", "category": "post_trained_general", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "MultiChallenge", "category": "post_trained_general", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "Arena-Hard-V2", "category": "post_trained_general", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "RULER-100", "category": "post_trained_long_context", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "AA-LCR", "category": "post_trained_long_context", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "MMLU-ProX", "category": "post_trained_multilingual", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}, {"name": "WMT24++", "category": "post_trained_multilingual", "models": ["nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-FP8"]}]}
{"id": "nemotron-orchestrator-paper", "type": "paper", "title": "ToolOrchestra", "url": "https://arxiv.org/abs/2511.21689", "arxiv_id": "2511.21689", "models": ["nvidia/Nemotron-Orchestrator-8B"], "notes": null, "benchmarks": [{"name": "HLE", "category": "agentic", "models": null}, {"name": "FRAMES", "category": "agentic", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}]}
{"id": "nemotron-terminal-paper", "type": "paper", "title": "On Data Engineering for Scaling LLM Terminal Capabilities", "url": "https://arxiv.org/abs/2602.21193", "arxiv_id": "2602.21193", "models": ["nvidia/Nemotron-Terminal-8B", "nvidia/Nemotron-Terminal-14B", "nvidia/Nemotron-Terminal-32B"], "notes": "AIME/LiveCodeBench/SWE-bench in the paper are used only to validate the teacher model (DeepSeek-V3.2), not to score the Nemotron-Terminal models themselves.", "benchmarks": [{"name": "Terminal-Bench 2.0", "category": "agentic", "models": null}]}
{"id": "allenai-tmax-paper", "type": "paper", "title": "TMax: A simple recipe for terminal agents", "url": "https://arxiv.org/abs/2606.23321", "arxiv_id": "2606.23321", "models": ["allenai/tmax-2b", "allenai/tmax-4b", "allenai/tmax-9b", "allenai/tmax-27b"], "notes": "Only the 9B size is additionally evaluated on SWE-Bench Verified, AIME 2024/2025, and alternate agent harnesses for generalization checks.", "benchmarks": [{"name": "Terminal-Bench 2.0", "category": "agentic_terminal", "models": null}, {"name": "Terminal-Bench 2.1", "category": "agentic_terminal", "models": null}, {"name": "Terminal-Bench Lite", "category": "agentic_terminal", "models": null}, {"name": "SWE-bench Verified", "category": "coding_9b_only", "models": null}, {"name": "AIME 2024", "category": "math_9b_only", "models": null}, {"name": "AIME 2025", "category": "math_9b_only", "models": null}]}
{"id": "gemma-3-paper", "type": "paper", "title": "Gemma 3 Technical Report", "url": "https://arxiv.org/abs/2503.19786", "arxiv_id": "2503.19786", "models": ["google/gemma-3-270m-it", "google/gemma-3-27b-it"], "notes": null, "benchmarks": [{"name": "HellaSwag", "category": "commonsense_reading", "models": null}, {"name": "BoolQ", "category": "commonsense_reading", "models": null}, {"name": "PIQA", "category": "commonsense_reading", "models": null}, {"name": "SocialIQA", "category": "commonsense_reading", "models": null}, {"name": "TriviaQA", "category": "commonsense_reading", "models": null}, {"name": "ARC-Challenge", "category": "commonsense_reading", "models": null}, {"name": "WinoGrande", "category": "commonsense_reading", "models": null}, {"name": "DROP", "category": "commonsense_reading", "models": null}, {"name": "MMLU", "category": "broad_reasoning", "models": null}, {"name": "AGIEval", "category": "broad_reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "broad_reasoning", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "MGSM", "category": "multilingual", "models": null}, {"name": "FLoRes-101", "category": "multilingual", "models": null}, {"name": "XQuAD", "category": "multilingual", "models": null}, {"name": "Global-MMLU", "category": "multilingual", "models": null}, {"name": "ECLeKTic", "category": "multilingual", "models": null}, {"name": "IndicGenBench", "category": "multilingual", "models": null}, {"name": "MMMU", "category": "vision", "models": null}, {"name": "ChartQA", "category": "vision", "models": null}, {"name": "InfographicVQA", "category": "vision", "models": null}, {"name": "TallyQA", "category": "vision", "models": null}, {"name": "SpatialSense", "category": "vision", "models": null}, {"name": "BLINK", "category": "vision", "models": null}, {"name": "MMLU-Pro", "category": "cited_no_arxiv_tag", "models": null}, {"name": "LiveCodeBench", "category": "cited_no_arxiv_tag", "models": null}, {"name": "SimpleQA", "category": "cited_no_arxiv_tag", "models": null}, {"name": "FACTS Grounding", "category": "cited_no_arxiv_tag", "models": null}, {"name": "HiddenMath", "category": "cited_no_arxiv_tag", "models": null}, {"name": "IFEval", "category": "cited_no_arxiv_tag", "models": null}]}
{"id": "llama-3-4", "type": "paper", "title": "The Llama 3 Herd of Models (Llama-3.1); Llama 4 Maverick has no formal Meta arXiv paper, benchmarks from HF card/blog", "url": "https://arxiv.org/abs/2407.21783", "arxiv_id": "2407.21783", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-4-Maverick-17B-128E-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"], "notes": "The 2204.05149 tag on both Hub repos is a carbon-footprint-methodology citation, not the technical report. Llama 3.1's real paper is 2407.21783; Llama 4 Maverick is blog/card-only.", "benchmarks": [{"name": "MMLU", "category": "llama3.1_knowledge", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "MMLU-Pro", "category": "llama3.1_knowledge", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "IFEval", "category": "llama3.1_knowledge", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "ARC-Challenge", "category": "llama3.1_knowledge", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "GPQA", "category": "llama3.1_knowledge", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "GSM8K", "category": "llama3.1_math", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "MATH", "category": "llama3.1_math", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "HumanEval", "category": "llama3.1_coding", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "MBPP+", "category": "llama3.1_coding", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "MultiPL-E", "category": "llama3.1_coding", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "API-Bank", "category": "llama3.1_agentic", "models": ["meta-llama/Llama-3.1-405B-Instruct"]}, {"name": "BFCL", "category": "llama3.1_agentic", "models": ["meta-llama/Llama-3.1-405B-Instruct"]}, {"name": "Gorilla API-Bench", "category": "llama3.1_agentic", "models": ["meta-llama/Llama-3.1-405B-Instruct"]}, {"name": "Nexus", "category": "llama3.1_agentic", "models": ["meta-llama/Llama-3.1-405B-Instruct"]}, {"name": "MGSM", "category": "llama3.1_multilingual", "models": ["meta-llama/Llama-3.1-405B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "meta-llama/Meta-Llama-3-8B-Instruct"]}, {"name": "MMLU", "category": "llama4_knowledge", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MMLU-Pro", "category": "llama4_knowledge", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "GPQA-Diamond", "category": "llama4_knowledge", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MATH", "category": "llama4_math", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MBPP", "category": "llama4_coding", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "LiveCodeBench", "category": "llama4_coding", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "TydiQA", "category": "llama4_multilingual", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MGSM", "category": "llama4_multilingual", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MTOB", "category": "llama4_long_context", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MMMU", "category": "llama4_vision", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MMMU-Pro", "category": "llama4_vision", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "MathVista", "category": "llama4_vision", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "ChartQA", "category": "llama4_vision", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "DocVQA", "category": "llama4_vision", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}, {"name": "LMArena", "category": "llama4_human_pref", "models": ["meta-llama/Llama-4-Maverick-17B-128E-Instruct"]}]}
{"id": "mirothinker-v1.5-paper", "type": "paper", "title": "MiroThinker v1.5 technical report", "url": "https://arxiv.org/abs/2511.11793", "arxiv_id": "2511.11793", "models": ["miromind-ai/MiroThinker-v1.5-235B", "miromind-ai/MiroThinker-v1.5-30B"], "notes": "No coding or math benchmarks used - purely an agentic search/deep-research model.", "benchmarks": [{"name": "BrowseComp", "category": "agentic_search", "models": null}, {"name": "BrowseComp-ZH", "category": "agentic_search", "models": null}, {"name": "WebWalkerQA", "category": "agentic_search", "models": null}, {"name": "xBench-DeepSearch", "category": "agentic_search", "models": null}, {"name": "FRAMES", "category": "agentic_search", "models": null}, {"name": "Seal-0", "category": "agentic_search", "models": null}, {"name": "GAIA", "category": "agentic_search", "models": null}, {"name": "HLE", "category": "agentic_search", "models": null}]}
{"id": "gpt-oss-report", "type": "report", "title": "gpt-oss Model Card", "url": "https://arxiv.org/abs/2508.10925", "arxiv_id": "2508.10925", "models": ["openai/gpt-oss-120b", "openai/gpt-oss-20b"], "notes": "Preparedness Framework (frontier-risk) evals were run on gpt-oss-120b only, not 20b.", "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMMLU", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "AIME 2024", "category": "math", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "Codeforces", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "Aider-Polyglot", "category": "coding", "models": null}, {"name": "TAU-Bench", "category": "agentic", "models": null}, {"name": "HealthBench", "category": "health", "models": null}, {"name": "HealthBench Hard", "category": "health", "models": null}, {"name": "HealthBench Consensus", "category": "health", "models": null}, {"name": "Disallowed Content Eval", "category": "safety", "models": null}, {"name": "StrongReject", "category": "safety", "models": null}, {"name": "SimpleQA", "category": "safety", "models": null}, {"name": "PersonQA", "category": "safety", "models": null}, {"name": "BBQ", "category": "safety", "models": null}, {"name": "Bio/chem risk sets", "category": "preparedness_120b_only", "models": null}, {"name": "CTF", "category": "preparedness_120b_only", "models": null}, {"name": "Cyber Range", "category": "preparedness_120b_only", "models": null}, {"name": "SWE-bench Verified", "category": "preparedness_120b_only", "models": null}, {"name": "PaperBench", "category": "preparedness_120b_only", "models": null}]}
{"id": "step-3.5-3.7-flash", "type": "paper", "title": "Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters (+ Step-3.7-Flash model card, no paper)", "url": "https://arxiv.org/abs/2602.10604", "arxiv_id": "2602.10604", "models": ["stepfun-ai/Step-3.5-Flash", "stepfun-ai/Step-3.7-Flash"], "notes": "Step-3.7-Flash has no arXiv paper; it's VLM-oriented and shares only partial overlap with 3.5-Flash's suite (Terminal-Bench, SWE-Bench, HLE).", "benchmarks": [{"name": "BBH", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MMLU-Redux", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MMLU-Pro", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "HellaSwag", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "WinoGrande", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "GPQA", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "SuperGPQA", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "SimpleQA", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "GSM8K", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MATH", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "HumanEval+", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MBPP+", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MultiPL-E", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "C-Eval", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "CMMLU", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "C-SimpleQA", "category": "step3.5_pretraining", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "HLE", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MMLU-Pro", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "GPQA-Diamond", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "AIME 2025", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "HMMT", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "IMO-AnswerBench", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "LiveCodeBench v6", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "SWE-bench Verified", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "SWE-bench Multilingual", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "Terminal-Bench 2.0", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "TAU2-Bench", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "GAIA", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "BrowseComp", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "BrowseComp-ZH", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "xBench-DeepSearch", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "Arena-Hard-V2", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "IFBench", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MultiChallenge", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "LongBench v2", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "MRCR", "category": "step3.5_post_training", "models": ["stepfun-ai/Step-3.5-Flash"]}, {"name": "SimpleVQA", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "V*", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "Claw-Eval", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "Toolathlon", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "HLE w/ Tools", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "SWE-bench Pro", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "Terminal-Bench 2.1", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}, {"name": "GDPval-AA", "category": "step3.7_vlm", "models": ["stepfun-ai/Step-3.7-Flash"]}]}
{"id": "minimax-m2.5-m2.7-cards", "type": "model_card", "title": "MiniMax-M2.5 / M2.7 model cards", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "arxiv_id": null, "models": ["MiniMaxAI/MiniMax-M2.5", "MiniMaxAI/MiniMax-M2.7"], "notes": null, "benchmarks": [{"name": "SWE-bench Verified", "category": "m2.5_coding", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "SWE-bench Multilingual", "category": "m2.5_coding", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "SWE-bench Pro", "category": "m2.5_coding", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "Multi-SWE-bench", "category": "m2.5_coding", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "VIBE-Pro", "category": "m2.5_coding", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "Terminal-Bench 2.0", "category": "m2.5_coding", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "AIME 2025", "category": "m2.5_reasoning", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "GPQA-Diamond", "category": "m2.5_reasoning", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "HLE", "category": "m2.5_reasoning", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "SciCode", "category": "m2.5_reasoning", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "IFBench", "category": "m2.5_reasoning", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "AA-LCR", "category": "m2.5_reasoning", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "BrowseComp", "category": "m2.5_search", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "WideSearch", "category": "m2.5_search", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "RISE", "category": "m2.5_search", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "GDPval-MM", "category": "m2.5_office", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "MEWC", "category": "m2.5_office", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "Finance Modeling", "category": "m2.5_office", "models": ["MiniMaxAI/MiniMax-M2.5"]}, {"name": "SWE-bench Pro", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "SWE-bench Multilingual", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "Multi-SWE-bench", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "VIBE-Pro", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "Terminal-Bench 2.0", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "NL2Repo", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "MLE-Bench-Lite", "category": "m2.7_coding", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "GDPval-AA", "category": "m2.7_office", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "Toolathlon", "category": "m2.7_office", "models": ["MiniMaxAI/MiniMax-M2.7"]}, {"name": "MM Claw", "category": "m2.7_office", "models": ["MiniMaxAI/MiniMax-M2.7"]}]}
{"id": "laguna-family-cards", "type": "model_card", "title": "Laguna family model cards (MuVeraAI + poolside)", "url": "https://huggingface.co/poolside/Laguna-XS.2", "arxiv_id": null, "models": ["MuVeraAI/Laguna-XS.2", "poolside/Laguna-M.1", "poolside/Laguna-XS-2.1", "poolside/Laguna-XS.2"], "notes": "MuVeraAI/Laguna-XS.2 is an unedited older repack of poolside/Laguna-XS.2's card (stale scores, leftover self-references to the poolside repo).", "benchmarks": [{"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding", "models": null}, {"name": "SWE-bench Pro", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}]}
{"id": "orionllm-cards", "type": "model_card", "title": "OrionLLM model cards", "url": "https://huggingface.co/OrionLLM/GRM-2.6-Plus", "arxiv_id": null, "models": ["OrionLLM/GRM-2.6-Plus", "OrionLLM/Terminus-Qwen3-8b"], "notes": null, "benchmarks": [{"name": "MMLU-Pro", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "MMLU-Redux", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "C-Eval", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "GPQA-Diamond", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "SuperGPQA", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "LiveCodeBench v6", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "HMMT Feb 2026", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "AIME 2026", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "SWE-bench Verified", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "SWE-bench Pro", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "Terminal-Bench 2.0", "category": "grm2.6plus", "models": ["OrionLLM/GRM-2.6-Plus"]}, {"name": "Terminal-Bench 2.0", "category": "terminus_qwen3_8b", "models": ["OrionLLM/Terminus-Qwen3-8b"]}, {"name": "SWE-bench Verified", "category": "terminus_qwen3_8b", "models": ["OrionLLM/Terminus-Qwen3-8b"]}]}
{"id": "obscure-singles-cards", "type": "model_card", "title": "Obscure single-model cards (CohereLabs, mindlab-research, internlm, HelpingAI)", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", "arxiv_id": null, "models": ["CohereLabs/North-Mini-Code-1.0", "mindlab-research/Macaron-V1-Preview-749B", "internlm/Intern-S2-Preview", "HelpingAI/Dhanishtha-2.0-0126"], "notes": "HelpingAI/Dhanishtha-2.0-0126's model card is an unfilled auto-generated stub - no benchmark data exists.", "benchmarks": [{"name": "SWE-bench Verified", "category": "north_mini_code", "models": ["CohereLabs/North-Mini-Code-1.0"]}, {"name": "SWE-bench Pro", "category": "north_mini_code", "models": ["CohereLabs/North-Mini-Code-1.0"]}, {"name": "Terminal-Bench 2.0", "category": "north_mini_code", "models": ["CohereLabs/North-Mini-Code-1.0"]}, {"name": "Terminal-Bench Hard", "category": "north_mini_code", "models": ["CohereLabs/North-Mini-Code-1.0"]}, {"name": "SciCode", "category": "north_mini_code", "models": ["CohereLabs/North-Mini-Code-1.0"]}, {"name": "LiveCodeBench v6", "category": "north_mini_code", "models": ["CohereLabs/North-Mini-Code-1.0"]}, {"name": "Macaron LivingBench", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "VITA-Bench", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "A2UI-Bench", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "PinchBench", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "TAU3-Bench", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "SWE-bench Verified", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "Terminal-Bench 2.0", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "AIME 2026", "category": "macaron_v1", "models": ["mindlab-research/Macaron-V1-Preview-749B"]}, {"name": "Biology-Instructions", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "MicroVQA", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "Mol-Instructions", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "MolecularIQ", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "SFE", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "XLRS-Bench", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "SciReasoner", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "SGI-Bench", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "FrontierScience-Olympiad", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "FrontierScience-Research", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "SciCode", "category": "intern_s2_scientific", "models": ["internlm/Intern-S2-Preview"]}, {"name": "MMLU-Pro", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "MMMU-Pro", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "HLE", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "IMO-AnswerBench", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "MathVision", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "HMMT 2026", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "ChartQAPro", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "SimpleQA Verified", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "LiveCodeBench Pro", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "IFBench", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "PinchBench", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}, {"name": "SWE-bench Verified", "category": "intern_s2_general", "models": ["internlm/Intern-S2-Preview"]}]}
{"id": "mimo-v2-family-cards", "type": "model_card", "title": "XiaomiMiMo V2 family model cards", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro", "arxiv_id": null, "models": ["XiaomiMiMo/MiMo-V2-Flash", "XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"], "notes": null, "benchmarks": [{"name": "BBH", "category": "shared_base", "models": null}, {"name": "MMLU", "category": "shared_base", "models": null}, {"name": "MMLU-Redux", "category": "shared_base", "models": null}, {"name": "MMLU-Pro", "category": "shared_base", "models": null}, {"name": "DROP", "category": "shared_base", "models": null}, {"name": "ARC-Challenge", "category": "shared_base", "models": null}, {"name": "HellaSwag", "category": "shared_base", "models": null}, {"name": "WinoGrande", "category": "shared_base", "models": null}, {"name": "TriviaQA", "category": "shared_base", "models": null}, {"name": "GPQA-Diamond", "category": "shared_base", "models": null}, {"name": "GSM8K", "category": "shared_base", "models": null}, {"name": "MATH", "category": "shared_base", "models": null}, {"name": "AIME 2024", "category": "shared_base", "models": null}, {"name": "AIME 2025", "category": "shared_base", "models": null}, {"name": "HumanEval+", "category": "shared_base", "models": null}, {"name": "MBPP+", "category": "shared_base", "models": null}, {"name": "LiveCodeBench v6", "category": "shared_base", "models": null}, {"name": "SWE-bench", "category": "shared_base", "models": null}, {"name": "C-Eval", "category": "shared_base", "models": null}, {"name": "CMMLU", "category": "shared_base", "models": null}, {"name": "Global-MMLU", "category": "shared_base", "models": null}, {"name": "HLE", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "HMMT", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "Arena-Hard", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "LongBench v2", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "MRCR", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "SWE-bench Verified", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "SWE-bench Multilingual", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "Terminal-Bench Hard", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "Terminal-Bench 2.0", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "BrowseComp", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "TAU2-Bench", "category": "flash_post_training", "models": ["XiaomiMiMo/MiMo-V2-Flash"]}, {"name": "CharXiv", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "MMMU-Pro", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "HR-Bench", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "OmniDocBench", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "Claw-Eval Multimodal", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "Video-MME", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "MiMo Coding Bench", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "SWE-bench Pro", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}, {"name": "GraphWalks", "category": "v2.5_omnimodal", "models": ["XiaomiMiMo/MiMo-V2.5", "XiaomiMiMo/MiMo-V2.5-Pro"]}]}
{"id": "ornith-1.0-family-cards", "type": "model_card", "title": "Ornith-1.0 family model cards", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B", "arxiv_id": null, "models": ["deepreinforce-ai/Ornith-1.0-9B", "deepreinforce-ai/Ornith-1.0-35B", "deepreinforce-ai/Ornith-1.0-397B"], "notes": "All 3 sizes share one 10-item agentic coding suite; no knowledge/math benchmarks used at all.", "benchmarks": [{"name": "Terminal-Bench 2.1", "category": "agentic_coding", "models": null}, {"name": "SWE-bench Verified", "category": "agentic_coding", "models": null}, {"name": "SWE-bench Pro", "category": "agentic_coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "agentic_coding", "models": null}, {"name": "NL2Repo", "category": "agentic_coding", "models": null}, {"name": "Claw-Eval", "category": "agentic_coding", "models": null}, {"name": "SWE Atlas", "category": "agentic_coding", "models": null}]}
{"id": "deepseek-v3.2-report", "type": "report", "title": "DeepSeek-V3.2 technical report (PDF hosted in HF repo)", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "arxiv_id": null, "models": ["deepseek-ai/DeepSeek-V3.2"], "notes": null, "benchmarks": [{"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "HMMT", "category": "math", "models": null}, {"name": "IMO-AnswerBench", "category": "math", "models": null}, {"name": "IMO 2025", "category": "math", "models": null}, {"name": "CMO 2025", "category": "math", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "Codeforces", "category": "coding", "models": null}, {"name": "Aider-Polyglot", "category": "coding", "models": null}, {"name": "IOI 2025", "category": "coding", "models": null}, {"name": "ICPC 2025", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "code_agent", "models": null}, {"name": "SWE-bench Multilingual", "category": "code_agent", "models": null}, {"name": "Terminal-Bench 2.0", "category": "code_agent", "models": null}, {"name": "BrowseComp", "category": "search_agent", "models": null}, {"name": "BrowseComp-ZH", "category": "search_agent", "models": null}, {"name": "TAU2-Bench", "category": "tool_use", "models": null}, {"name": "MCP-Universe", "category": "tool_use", "models": null}, {"name": "MCP-Mark", "category": "tool_use", "models": null}, {"name": "Tool-Decathlon", "category": "tool_use", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "Fiction.liveBench", "category": "long_context", "models": null}]}
{"id": "ibm-granite-4.0-4.1-cards", "type": "model_card", "title": "IBM Granite 4.0/4.1 model cards", "url": "https://huggingface.co/ibm-granite/granite-4.1-8b", "arxiv_id": null, "models": ["ibm-granite/granite-4.0-h-350m", "ibm-granite/granite-4.0-h-tiny", "ibm-granite/granite-4.1-8b"], "notes": "4.1-8B additionally adds SimpleQA, MTBench, aggregated Eval+, MULTIPLE, Tulu3 Safety Eval vs. the 4.0-h generation.", "benchmarks": [{"name": "MMLU", "category": "shared_general", "models": null}, {"name": "MMLU-Pro", "category": "shared_general", "models": null}, {"name": "BBH", "category": "shared_general", "models": null}, {"name": "AGIEval", "category": "shared_general", "models": null}, {"name": "GPQA", "category": "shared_general", "models": null}, {"name": "AlpacaEval 2.0", "category": "shared_alignment", "models": null}, {"name": "IFEval", "category": "shared_alignment", "models": null}, {"name": "Arena-Hard", "category": "shared_alignment", "models": null}, {"name": "GSM8K", "category": "shared_math", "models": null}, {"name": "Minerva Math", "category": "shared_math", "models": null}, {"name": "DeepMind Math", "category": "shared_math", "models": null}, {"name": "HumanEval", "category": "shared_code", "models": null}, {"name": "HumanEval+", "category": "shared_code", "models": null}, {"name": "MBPP", "category": "shared_code", "models": null}, {"name": "MBPP+", "category": "shared_code", "models": null}, {"name": "CRUXEval-O", "category": "shared_code", "models": null}, {"name": "BigCodeBench", "category": "shared_code", "models": null}, {"name": "BFCL v3", "category": "shared_tool_calling", "models": null}, {"name": "MMMLU", "category": "shared_multilingual", "models": null}, {"name": "INCLUDE", "category": "shared_multilingual", "models": null}, {"name": "MGSM", "category": "shared_multilingual", "models": null}, {"name": "SALAD-Bench", "category": "shared_safety", "models": null}, {"name": "AttaQ", "category": "shared_safety", "models": null}, {"name": "SimpleQA", "category": "granite_4.1_additions", "models": ["ibm-granite/granite-4.1-8b"]}, {"name": "MT-Bench", "category": "granite_4.1_additions", "models": ["ibm-granite/granite-4.1-8b"]}, {"name": "MultiPL-E", "category": "granite_4.1_additions", "models": ["ibm-granite/granite-4.1-8b"]}, {"name": "Tulu3 Safety Eval", "category": "granite_4.1_additions", "models": ["ibm-granite/granite-4.1-8b"]}]}
{"id": "qwen3.5-3.6-smaller-sizes", "type": "model_card", "title": "Qwen3.5/3.6 smaller-size model cards (full benchmark tables, per model)", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "arxiv_id": null, "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B", "Qwen/Qwen3.5-4B", "Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B", "Qwen/Qwen3.6-35B-A3B"], "notes": "Full tables extracted directly from each model's own card (not diffed against the flagship). Qwen3.5-27B and 35B-A3B share a byte-identical table; Qwen3.5-0.8B and 2B likewise share one. Qwen3.5-4B and Qwen3.6-35B-A3B each have their own distinct suite.", "benchmarks": [{"name": "MMLU-Pro", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMLU-Redux", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "C-Eval", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "SuperGPQA", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "IFEval", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "IFBench", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MultiChallenge", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "HLE", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "GPQA-Diamond", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "HMMT Feb 2025", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "HMMT Nov 2025", "category": "knowledge", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "AA-LCR", "category": "long_context", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "LongBench v2", "category": "long_context", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "SWE-bench Verified", "category": "coding", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "LiveCodeBench v6", "category": "coding", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Codeforces", "category": "coding", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "OJBench", "category": "coding", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "FullStackBench", "category": "coding", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "BFCL-V4", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "TAU2-Bench", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "VITA-Bench", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "DeepPlanning", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "HLE w/ Tools", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "BrowseComp", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "BrowseComp-ZH", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "WideSearch", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Seal-0", "category": "agentic", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMMLU", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMLU-ProX", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "NOVA-63", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "INCLUDE", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Global PIQA", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "PolyMATH", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "WMT24++", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MAXIFE", "category": "multilingual", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMMU", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMMU-Pro", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MathVision", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MathVista", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "DynaMath", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "ZeroBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "ZeroBench_sub", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "VlmsAreBlind", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "BabyVision", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "RealWorldQA", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMStar", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "SimpleVQA", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "HallusionBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "OmniDocBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "CharXiv", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMLongBench-Doc", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "CC-OCR", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "AI2D", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "OCRBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "ERQA", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "CountBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "RefCOCO", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "ODinW13", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "EmbSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "RefSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "LingoQA", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Hypersim", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "SUNRGBD", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Nuscenes", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "Video-MME", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "VideoMMMU", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MLVU", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MVBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "LVBench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMVU", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "ScreenSpot Pro", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "OSWorld-Verified", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "AndroidWorld", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "TIR-Bench", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "V*", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "SLAKE", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "PMC-VQA", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MedXpertQA-MM", "category": "vision", "models": ["Qwen/Qwen3.5-27B", "Qwen/Qwen3.5-35B-A3B"]}, {"name": "MMLU-Pro", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMLU-Redux", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "C-Eval", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "SuperGPQA", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "GPQA-Diamond", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "IFEval", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "IFBench", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MultiChallenge", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "HMMT Feb 2025", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "HMMT Nov 2025", "category": "knowledge", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "AA-LCR", "category": "long_context", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "LongBench v2", "category": "long_context", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "LiveCodeBench v6", "category": "coding", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "OJBench", "category": "coding", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "BFCL-V4", "category": "agentic", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "TAU2-Bench", "category": "agentic", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "VITA-Bench", "category": "agentic", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "DeepPlanning", "category": "agentic", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMMLU", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMLU-ProX", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "NOVA-63", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "INCLUDE", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "Global PIQA", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "PolyMATH", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "WMT24++", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MAXIFE", "category": "multilingual", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMMU", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMMU-Pro", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MathVision", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MathVista", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "We-Math", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "DynaMath", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "ZeroBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "ZeroBench_sub", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "VlmsAreBlind", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "BabyVision", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "RealWorldQA", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMStar", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "SimpleVQA", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "HallusionBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "OmniDocBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "CharXiv", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMLongBench-Doc", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "CC-OCR", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "AI2D", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "OCRBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "ERQA", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "CountBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "RefCOCO", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "EmbSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "RefSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "LingoQA", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "Hypersim", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "Nuscenes", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "Video-MME", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "VideoMMMU", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MLVU", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MVBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "LVBench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMVU", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "ScreenSpot Pro", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "OSWorld-Verified", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "AndroidWorld", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "TIR-Bench", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "V*", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "SLAKE", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "PMC-VQA", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MedXpertQA-MM", "category": "vision", "models": ["Qwen/Qwen3.5-4B"]}, {"name": "MMLU-Pro", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMLU-Redux", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "C-Eval", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "SuperGPQA", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "GPQA", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "IFEval", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "IFBench", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MultiChallenge", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "HMMT Feb 2025", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "HMMT Nov 2025", "category": "knowledge", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "AA-LCR", "category": "long_context", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "LongBench v2", "category": "long_context", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "BFCL-V4", "category": "agentic", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "TAU2-Bench", "category": "agentic", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMMLU", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMLU-ProX", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "NOVA-63", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "INCLUDE", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "Global PIQA", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "PolyMATH", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "WMT24++", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MAXIFE", "category": "multilingual", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMMU", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMMU-Pro", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MathVision", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MathVista", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "DynaMath", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "ZeroBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "ZeroBench_sub", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "VlmsAreBlind", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "RealWorldQA", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMStar", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "SimpleVQA", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "HallusionBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMLongBench-Doc", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "AI2D", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "CC-OCR", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "OmniDocBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "CharXiv", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "OCRBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "RefCOCO", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "CountBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "ODinW13", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "ERQA", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "EmbSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "RefSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "Hypersim", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "SUNRGBD", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "Nuscenes", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "Video-MME", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "VideoMMMU", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MLVU", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MVBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "LVBench", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MMVU", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "ScreenSpot Pro", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "SLAKE", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "PMC-VQA", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "MedXpertQA-MM", "category": "vision", "models": ["Qwen/Qwen3.5-0.8B", "Qwen/Qwen3.5-2B"]}, {"name": "SWE-bench Verified", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "SWE-bench Multilingual", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "SWE-bench Pro", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "Claw-Eval", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "SkillsBench", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "QwenClawBench", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "NL2Repo", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "QwenWebBench", "category": "coding", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "TAU3-Bench", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "VITA-Bench", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "DeepPlanning", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "Tool-Decathlon", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MCP-Mark", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MCP-Atlas", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "WideSearch", "category": "agentic", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MMLU-Pro", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MMLU-Redux", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "SuperGPQA", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "C-Eval", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "GPQA", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "HLE", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "LiveCodeBench v6", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "HMMT Feb 2025", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "HMMT Nov 2025", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "HMMT Feb 2026", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "IMO-AnswerBench", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "AIME 2026", "category": "knowledge", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MMMU", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MMMU-Pro", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MathVista", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "ZeroBench_sub", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "RealWorldQA", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MMBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "SimpleVQA", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "HallusionBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "OmniDocBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "CharXiv", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "CC-OCR", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "AI2D", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "RefCOCO", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "ODinW13", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "EmbSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "RefSpatialBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "Video-MME", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "VideoMMMU", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MLVU", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "MVBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}, {"name": "LVBench", "category": "vision", "models": ["Qwen/Qwen3.6-35B-A3B"]}]}
{"id": "kimi-k2-original-paper", "type": "paper", "title": "Kimi K2: Open Agentic Intelligence", "url": "https://arxiv.org/abs/2507.20534", "arxiv_id": "2507.20534", "models": ["moonshotai/Kimi-K2-Instruct", "moonshotai/Kimi-K2-Thinking"], "notes": "Kimi-K2-Thinking additions (IMO-AnswerBench, HealthBench, BrowseComp, BrowseComp-ZH, Seal-0, FinSearchComp-T3, SciCode, Longform Writing) come from its own HF model card, not the original paper.", "benchmarks": [{"name": "MMLU", "category": "base", "models": null}, {"name": "MMLU-Pro", "category": "base", "models": null}, {"name": "MMLU-Redux", "category": "base", "models": null}, {"name": "BBH", "category": "base", "models": null}, {"name": "TriviaQA", "category": "base", "models": null}, {"name": "SuperGPQA", "category": "base", "models": null}, {"name": "SimpleQA", "category": "base", "models": null}, {"name": "HellaSwag", "category": "base", "models": null}, {"name": "AGIEval", "category": "base", "models": null}, {"name": "GPQA-Diamond", "category": "base", "models": null}, {"name": "ARC-Challenge", "category": "base", "models": null}, {"name": "WinoGrande", "category": "base", "models": null}, {"name": "EvalPlus", "category": "base", "models": null}, {"name": "LiveCodeBench v6", "category": "base", "models": null}, {"name": "CRUXEval", "category": "base", "models": null}, {"name": "GSM8K", "category": "base", "models": null}, {"name": "MATH", "category": "base", "models": null}, {"name": "C-Eval", "category": "base", "models": null}, {"name": "CMMLU", "category": "base", "models": null}, {"name": "LiveCodeBench v6", "category": "instruct_coding", "models": null}, {"name": "OJBench", "category": "instruct_coding", "models": null}, {"name": "MultiPL-E", "category": "instruct_coding", "models": null}, {"name": "SWE-bench Verified", "category": "instruct_coding", "models": null}, {"name": "Terminal-Bench", "category": "instruct_coding", "models": null}, {"name": "Multi-SWE-bench", "category": "instruct_coding", "models": null}, {"name": "SWE-Lancer", "category": "instruct_coding", "models": null}, {"name": "PaperBench", "category": "instruct_coding", "models": null}, {"name": "Aider-Polyglot", "category": "instruct_coding", "models": null}, {"name": "TAU2-Bench", "category": "instruct_agentic", "models": null}, {"name": "ACEBench", "category": "instruct_agentic", "models": null}, {"name": "AIME 2024", "category": "instruct_math_science_logic", "models": null}, {"name": "AIME 2025", "category": "instruct_math_science_logic", "models": null}, {"name": "MATH-500", "category": "instruct_math_science_logic", "models": null}, {"name": "HMMT 2025", "category": "instruct_math_science_logic", "models": null}, {"name": "ZebraLogic", "category": "instruct_math_science_logic", "models": null}, {"name": "AutoLogi", "category": "instruct_math_science_logic", "models": null}, {"name": "GPQA-Diamond", "category": "instruct_math_science_logic", "models": null}, {"name": "HLE", "category": "instruct_math_science_logic", "models": null}, {"name": "MRCR", "category": "instruct_long_context", "models": null}, {"name": "DROP", "category": "instruct_long_context", "models": null}, {"name": "FRAMES", "category": "instruct_long_context", "models": null}, {"name": "LongBench v2", "category": "instruct_long_context", "models": null}, {"name": "FACTS Grounding", "category": "instruct_factuality", "models": null}, {"name": "IFEval", "category": "instruct_general", "models": null}, {"name": "MultiChallenge", "category": "instruct_general", "models": null}, {"name": "SimpleQA", "category": "instruct_general", "models": null}, {"name": "LiveBench", "category": "instruct_general", "models": null}, {"name": "Arena-Hard-Auto", "category": "instruct_general", "models": null}, {"name": "IMO-AnswerBench", "category": "thinking_additions", "models": null}, {"name": "HealthBench", "category": "thinking_additions", "models": null}, {"name": "BrowseComp", "category": "thinking_additions", "models": null}, {"name": "BrowseComp-ZH", "category": "thinking_additions", "models": null}, {"name": "Seal-0", "category": "thinking_additions", "models": null}, {"name": "FinSearchComp-T3", "category": "thinking_additions", "models": null}, {"name": "SciCode", "category": "thinking_additions", "models": null}, {"name": "Longform Writing", "category": "thinking_additions", "models": null}]}
{"id": "claude-5-family-system-cards", "type": "report", "title": "Anthropic system cards for Claude Opus 4.8, Sonnet 5, Haiku 4.5, and Fable 5", "url": "https://anthropic.com/system-cards", "arxiv_id": null, "models": ["claude-opus-4-8", "claude-sonnet-5", "claude-haiku-4-5", "claude-fable-5"], "notes": "Each model has its own system card (not one shared document) but the suites overlap heavily. Not all benchmarks are run on every tier — frontier-only evals (USAMO, GDPval-AA, BioMysteryBench, ExploitBench, domain-specific agentic benchmarks) cluster on Opus/Fable and are largely skipped for Haiku. Not on any HF leaderboard (closed API models); scores were cross-referenced partly via other labs' model cards that cite Claude comparison numbers.", "benchmarks": [{"name": "SWE-bench Verified", "category": "coding_agentic", "models": null}, {"name": "SWE-bench Pro", "category": "coding_agentic", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding_agentic", "models": null}, {"name": "CursorBench", "category": "coding_agentic", "models": null}, {"name": "FrontierCode", "category": "coding_agentic", "models": null}, {"name": "OSWorld-Verified", "category": "agentic_computer_use", "models": null}, {"name": "Online-Mind2Web", "category": "agentic_computer_use", "models": null}, {"name": "BrowseComp", "category": "agentic_computer_use", "models": null}, {"name": "MCP-Atlas", "category": "agentic_computer_use", "models": null}, {"name": "TAU2-Bench", "category": "agentic_computer_use", "models": null}, {"name": "Legal Agent Benchmark", "category": "agentic_domain_specific_opus_only", "models": null}, {"name": "Finance Agent v2", "category": "agentic_domain_specific_opus_only", "models": null}, {"name": "CoCounsel Legal", "category": "agentic_domain_specific_opus_only", "models": null}, {"name": "USAMO", "category": "math_reasoning", "models": null}, {"name": "HLE", "category": "math_reasoning", "models": null}, {"name": "HLE w/ Tools", "category": "math_reasoning", "models": null}, {"name": "AIME 2025", "category": "math_reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMMLU", "category": "knowledge", "models": null}, {"name": "GDPval-AA", "category": "knowledge", "models": null}, {"name": "GDPval (vision, no tools)", "category": "vision", "models": null}, {"name": "Automated Behavioral Audit", "category": "safety_alignment", "models": null}, {"name": "BioMysteryBench", "category": "safety_alignment", "models": null}, {"name": "ExploitBench", "category": "safety_alignment", "models": null}, {"name": "Firefox Exploit Development", "category": "safety_alignment", "models": null}, {"name": "OSS-Fuzz", "category": "safety_alignment", "models": null}, {"name": "CyScenarioBench", "category": "safety_alignment", "models": null}, {"name": "CyberGym", "category": "safety_alignment", "models": null}, {"name": "Over-refusal rate", "category": "safety_alignment", "models": null}]}
{"id": "gpt-5.5-system-card-and-blog", "type": "report", "title": "GPT-5.5 System Card + \"Introducing GPT-5.5\" announcement blog", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "arxiv_id": null, "models": ["gpt-5.5"], "notes": "Released April 23, 2026. Capability numbers from the announcement blog comparison chart (vs GPT-5.4, Claude, Gemini 3.1 Pro); safety/preparedness from the official system card PDF.", "benchmarks": [{"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "FrontierMath Tier 1-3", "category": "math", "models": null}, {"name": "FrontierMath Tier 4", "category": "math", "models": null}, {"name": "Terminal-Bench 2.0", "category": "coding", "models": null}, {"name": "SWE-bench Pro", "category": "coding", "models": null}, {"name": "Expert-SWE", "category": "coding", "models": null}, {"name": "GDPval", "category": "agentic", "models": null}, {"name": "OSWorld-Verified", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}, {"name": "MCP-Atlas", "category": "agentic", "models": null}, {"name": "ARC-AGI-2", "category": "abstraction_reasoning", "models": null}, {"name": "MRCR v2", "category": "long_context", "models": null}, {"name": "CyberGym", "category": "cybersecurity_capability", "models": null}, {"name": "Production Benchmarks (disallowed content)", "category": "safety_disallowed_content", "models": null}, {"name": "Image-input disallowed-content eval", "category": "safety_vision", "models": null}, {"name": "Destructive-action avoidance eval", "category": "safety_agentic", "models": null}, {"name": "User-confirmation eval", "category": "safety_agentic", "models": null}, {"name": "Multiturn jailbreak eval", "category": "safety_robustness", "models": null}, {"name": "Prompt-injection-in-connectors eval", "category": "safety_robustness", "models": null}, {"name": "HealthBench", "category": "health", "models": null}, {"name": "HealthBench Hard", "category": "health", "models": null}, {"name": "HealthBench Consensus", "category": "health", "models": null}, {"name": "HealthBench Professional", "category": "health", "models": null}, {"name": "Mental health dynamic benchmark", "category": "mental_health", "models": null}, {"name": "Emotional reliance benchmark", "category": "mental_health", "models": null}, {"name": "Self-harm dynamic benchmark", "category": "mental_health", "models": null}, {"name": "User-flagged hallucination eval", "category": "hallucination", "models": null}, {"name": "CoT Monitorability suite", "category": "alignment", "models": null}, {"name": "CoT Controllability (CoT-Control)", "category": "alignment", "models": null}, {"name": "First-Person Fairness Evaluation", "category": "bias", "models": null}, {"name": "Multimodal Troubleshooting Virology", "category": "preparedness_bio_chem", "models": null}, {"name": "ProtocolQA Open-Ended", "category": "preparedness_bio_chem", "models": null}, {"name": "Tacit Knowledge and Troubleshooting", "category": "preparedness_bio_chem", "models": null}, {"name": "TroubleshootingBench", "category": "preparedness_bio_chem", "models": null}, {"name": "Bio Bug Bounty", "category": "preparedness_bio_chem", "models": null}, {"name": "Capture the Flag (Professional)", "category": "preparedness_cyber", "models": null}, {"name": "CVE-Bench", "category": "preparedness_cyber", "models": null}, {"name": "Cyber Range", "category": "preparedness_cyber", "models": null}, {"name": "VulnLMP", "category": "preparedness_cyber", "models": null}, {"name": "Monorepo-Bench", "category": "preparedness_self_improvement", "models": null}, {"name": "MLE-Bench", "category": "preparedness_self_improvement", "models": null}, {"name": "Internal Research Debugging Eval", "category": "preparedness_self_improvement", "models": null}, {"name": "OpenAI-Proof Q&A", "category": "preparedness_self_improvement", "models": null}, {"name": "Apollo Research eval suite", "category": "sandbagging_scheming", "models": null}]}
{"id": "gpt-5.6-preview-system-card", "type": "report", "title": "GPT-5.6 Preview (Sol/Terra/Luna) System Card + secondary coverage", "url": "https://deploymentsafety.openai.com/gpt-5-6-preview/gpt-5-6-preview.pdf", "arxiv_id": null, "models": ["gpt-5.6"], "notes": "Limited preview as of June 2026 (codenamed Sol/Terra/Luna), no full ChatGPT release yet. Capability numbers (Terminal-Bench 2.1, Agent's Last Exam) are OpenAI-published; SWE-bench Pro was notably NOT published at preview; other capability figures circulating (FrontierMath T4, SWE-bench Verified) are explicitly unconfirmed third-party estimates, excluded here. Safety suite mostly retains GPT-5.5's evals with several additions/replacements.", "benchmarks": [{"name": "Terminal-Bench 2.1", "category": "coding", "models": null}, {"name": "Agent's Last Exam", "category": "agentic", "models": null}, {"name": "ExploitBench", "category": "cybersecurity_capability", "models": null}, {"name": "Production Benchmarks (disallowed content, reorganized categories)", "category": "safety_disallowed_content", "models": null}, {"name": "Multiturn jailbreak eval", "category": "safety_robustness", "models": null}, {"name": "Prompt-injection-in-connectors eval", "category": "safety_robustness", "models": null}, {"name": "Search and Function-Calling prompt-injection eval", "category": "safety_robustness", "models": null}, {"name": "CoT Monitorability suite", "category": "alignment", "models": null}, {"name": "CoT Controllability (CoT-Control)", "category": "alignment", "models": null}, {"name": "Metagaming evaluation", "category": "alignment", "models": null}, {"name": "Multimodal Troubleshooting Virology", "category": "preparedness_bio_chem", "models": null}, {"name": "ProtocolQA Open-Ended", "category": "preparedness_bio_chem", "models": null}, {"name": "Tacit Knowledge and Troubleshooting", "category": "preparedness_bio_chem", "models": null}, {"name": "TroubleshootingBench", "category": "preparedness_bio_chem", "models": null}, {"name": "AAV Capsid Packaging Prediction", "category": "preparedness_bio_chem", "models": null}, {"name": "Capture the Flag (Professional)", "category": "preparedness_cyber", "models": null}, {"name": "CVE-Bench", "category": "preparedness_cyber", "models": null}, {"name": "Cyber Range", "category": "preparedness_cyber", "models": null}, {"name": "VulnLMP", "category": "preparedness_cyber", "models": null}, {"name": "ExploitGym", "category": "preparedness_cyber", "models": null}, {"name": "SEC-Bench Pro", "category": "preparedness_cyber", "models": null}, {"name": "KernelGen 1P", "category": "preparedness_self_improvement", "models": null}, {"name": "NanoGPT", "category": "preparedness_self_improvement", "models": null}, {"name": "PostTrainBench Lite", "category": "preparedness_self_improvement", "models": null}, {"name": "MLE-Bench", "category": "preparedness_self_improvement", "models": null}, {"name": "Internal Research Debugging Eval", "category": "preparedness_self_improvement", "models": null}, {"name": "METR Time Horizon", "category": "preparedness_self_improvement", "models": null}, {"name": "Apollo Research eval suite", "category": "sandbagging_scheming", "models": null}]}
{"id": "gemini-2.5-paper", "type": "paper", "title": "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities", "url": "https://arxiv.org/abs/2507.06261", "arxiv_id": "2507.06261", "models": ["gemini-2.5-pro", "gemini-2.5-flash", "gemini-2.0-flash", "gemini-2.0-flash-lite"], "notes": "Not on any HF leaderboard (closed API models). Paper also compares against Gemini 1.5 Flash/Pro as legacy baselines (not added as separate model entries here). HiddenMath-Hard, Global-MMLU-Lite, ECLeKTic, ZeroBench, Vibe-Eval, and BetterChartQA are evaluated on Gemini models only, not run on competitor models in the paper's comparison tables. Section 2's HTML rendering was empty on arXiv; benchmarks extracted from the PDF tables.", "benchmarks": [{"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "Aider-Polyglot", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning_knowledge", "models": null}, {"name": "HLE", "category": "reasoning_knowledge", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "HiddenMath-Hard", "category": "math", "models": null}, {"name": "SimpleQA", "category": "factuality", "models": null}, {"name": "FACTS Grounding", "category": "factuality", "models": null}, {"name": "Global-MMLU-Lite", "category": "multilingual", "models": null}, {"name": "ECLeKTic", "category": "multilingual", "models": null}, {"name": "LOFT", "category": "long_context", "models": null}, {"name": "MRCR v2", "category": "long_context", "models": null}, {"name": "MMMU", "category": "vision_multimodal", "models": null}, {"name": "Vibe-Eval", "category": "vision_multimodal", "models": null}, {"name": "ZeroBench", "category": "vision_multimodal", "models": null}, {"name": "BetterChartQA", "category": "vision_multimodal", "models": null}, {"name": "FLEURS", "category": "audio", "models": null}, {"name": "CoVoST", "category": "audio", "models": null}, {"name": "ActivityNet-QA", "category": "video", "models": null}, {"name": "EgoTempo", "category": "video", "models": null}, {"name": "Perception Test", "category": "video", "models": null}, {"name": "QVHighlights", "category": "video", "models": null}, {"name": "VideoMMMU", "category": "video", "models": null}, {"name": "1H-VideoQA", "category": "video", "models": null}, {"name": "LVBench", "category": "video", "models": null}, {"name": "Video-MME", "category": "video", "models": null}, {"name": "VATEX", "category": "video", "models": null}, {"name": "VATEX-ZH", "category": "video", "models": null}, {"name": "YouCook2 Cap", "category": "video", "models": null}, {"name": "Minerva (video reasoning)", "category": "video", "models": null}, {"name": "Neptune", "category": "video", "models": null}]}
{"id": "hy3-full-release-card", "type": "model_card", "title": "tencent/Hy3 full-release model card (chart images, no arXiv paper)", "url": "https://huggingface.co/tencent/Hy3", "arxiv_id": null, "models": ["tencent/Hy3"], "notes": "Full/final release, distinct from tencent/Hy3-preview (already in this DB under hy3-preview-card). Benchmarks come entirely from two chart images (assets/benchmark.png, assets/benchmark-appendix.png), no text table or paper. Roughly 2x broader suite than the preview: retains most preview benchmarks (with version bumps, e.g. Terminal-Bench 2.0->2.1) while dropping Tsinghua Qiuzhen Math PhD Exam, CHSBO 2025, AdvancedIF, and LongBench v2 from the preview suite, and adding many new public + internal evals.", "benchmarks": [{"name": "SWE-bench Verified", "category": "agentic_coding", "models": null}, {"name": "SWE-bench Multilingual", "category": "agentic_coding", "models": null}, {"name": "SWE-bench Pro", "category": "agentic_coding", "models": null}, {"name": "Terminal-Bench 2.1", "category": "agentic_coding", "models": null}, {"name": "NL2Repo", "category": "agentic_coding", "models": null}, {"name": "DeepSWE", "category": "agentic_coding", "models": null}, {"name": "Hy-Backend 2.0", "category": "agentic_coding", "models": null}, {"name": "Hy-SWE Max", "category": "agentic_coding", "models": null}, {"name": "Hy-CompanyBench", "category": "agentic_coding", "models": null}, {"name": "BrowseComp", "category": "agentic_search", "models": null}, {"name": "WideSearch", "category": "agentic_search", "models": null}, {"name": "DeepSearchQA", "category": "agentic_search", "models": null}, {"name": "MCP-Atlas", "category": "working_agent_tool_use", "models": null}, {"name": "Toolathlon", "category": "working_agent_tool_use", "models": null}, {"name": "Apex-Agent", "category": "working_agent_tool_use", "models": null}, {"name": "Claw-Eval", "category": "working_agent_tool_use", "models": null}, {"name": "WildClawBench", "category": "working_agent_tool_use", "models": null}, {"name": "SkillsBench", "category": "working_agent_tool_use", "models": null}, {"name": "e-bench", "category": "working_agent_tool_use", "models": null}, {"name": "Hy-FinModelBench", "category": "working_agent_tool_use", "models": null}, {"name": "ProdBench", "category": "working_agent_tool_use", "models": null}, {"name": "Hy-SkillsWorld", "category": "working_agent_tool_use", "models": null}, {"name": "HLE w/ Tools", "category": "stem_agent", "models": null}, {"name": "Hy-Euler Pro", "category": "stem_agent", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning_math_science", "models": null}, {"name": "HLE", "category": "reasoning_math_science", "models": null}, {"name": "FrontierScience-Research", "category": "reasoning_math_science", "models": null}, {"name": "FrontierScience-Olympiad", "category": "reasoning_math_science", "models": null}, {"name": "USAMO 2026", "category": "reasoning_math_science", "models": null}, {"name": "MathArena Apex", "category": "reasoning_math_science", "models": null}, {"name": "ArxivMath", "category": "reasoning_math_science", "models": null}, {"name": "HorizonMath", "category": "reasoning_math_science", "models": null}, {"name": "Hy-Math", "category": "reasoning_math_science", "models": null}, {"name": "PHYBench", "category": "reasoning_math_science", "models": null}, {"name": "CMT-Benchmark", "category": "reasoning_math_science", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning_math_science", "models": null}, {"name": "SuperChem", "category": "reasoning_math_science", "models": null}, {"name": "CL-bench", "category": "long_context", "models": null}, {"name": "CL-bench life", "category": "long_context", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}]}
{"id": "deepseek-r1-paper", "type": "paper", "title": "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning", "url": "https://arxiv.org/abs/2501.12948", "arxiv_id": "2501.12948", "models": ["deepseek-ai/DeepSeek-R1"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMLU-Redux", "category": "knowledge", "models": null}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "C-Eval", "category": "knowledge", "models": null}, {"name": "CMMLU", "category": "knowledge", "models": null}, {"name": "CLUEWSC", "category": "knowledge", "models": null}, {"name": "FRAMES", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "SimpleQA", "category": "knowledge", "models": null}, {"name": "C-SimpleQA", "category": "knowledge", "models": null}, {"name": "MATH-500", "category": "math", "models": null}, {"name": "AIME 2024", "category": "math", "models": null}, {"name": "CNMO 2024", "category": "math", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "Codeforces", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "Aider-Polyglot", "category": "coding", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "AlpacaEval 2.0", "category": "general", "models": null}, {"name": "Arena-Hard", "category": "general", "models": null}, {"name": "Internal safety/jailbreak suite (unnamed individually)", "category": "safety", "models": null}]}
{"id": "llama-2-paper", "type": "paper", "title": "Llama 2: Open Foundation and Fine-Tuned Chat Models", "url": "https://arxiv.org/abs/2307.09288", "arxiv_id": "2307.09288", "models": ["meta-llama/Llama-2-7b-chat-hf"], "notes": null, "benchmarks": [{"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "PIQA", "category": "commonsense", "models": null}, {"name": "SIQA", "category": "commonsense", "models": null}, {"name": "HellaSwag", "category": "commonsense", "models": null}, {"name": "WinoGrande", "category": "commonsense", "models": null}, {"name": "ARC-Easy", "category": "commonsense", "models": null}, {"name": "ARC-Challenge", "category": "commonsense", "models": null}, {"name": "OpenBookQA", "category": "commonsense", "models": null}, {"name": "CommonsenseQA", "category": "commonsense", "models": null}, {"name": "Natural Questions", "category": "knowledge", "models": null}, {"name": "TriviaQA", "category": "knowledge", "models": null}, {"name": "SQuAD", "category": "reading_comprehension", "models": null}, {"name": "QuAC", "category": "reading_comprehension", "models": null}, {"name": "BoolQ", "category": "reading_comprehension", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "MMLU", "category": "aggregate", "models": null}, {"name": "BBH", "category": "aggregate", "models": null}, {"name": "AGIEval", "category": "aggregate", "models": null}, {"name": "TruthfulQA", "category": "safety", "models": null}, {"name": "ToxiGen", "category": "safety", "models": null}, {"name": "BOLD", "category": "safety", "models": null}]}
{"id": "mistral-7b-paper", "type": "paper", "title": "Mistral 7B", "url": "https://arxiv.org/abs/2310.06825", "arxiv_id": "2310.06825", "models": ["mistralai/Mistral-7B-Instruct-v0.2"], "notes": null, "benchmarks": [{"name": "HellaSwag", "category": "commonsense", "models": null}, {"name": "WinoGrande", "category": "commonsense", "models": null}, {"name": "PIQA", "category": "commonsense", "models": null}, {"name": "SIQA", "category": "commonsense", "models": null}, {"name": "OpenBookQA", "category": "commonsense", "models": null}, {"name": "ARC-Easy", "category": "commonsense", "models": null}, {"name": "ARC-Challenge", "category": "commonsense", "models": null}, {"name": "CommonsenseQA", "category": "commonsense", "models": null}, {"name": "Natural Questions", "category": "knowledge", "models": null}, {"name": "TriviaQA", "category": "knowledge", "models": null}, {"name": "BoolQ", "category": "reading_comprehension", "models": null}, {"name": "QuAC", "category": "reading_comprehension", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "MMLU", "category": "aggregate", "models": null}, {"name": "BBH", "category": "aggregate", "models": null}, {"name": "AGIEval", "category": "aggregate", "models": null}, {"name": "MT-Bench", "category": "general", "models": null}, {"name": "Internal refusal-rate / content-moderation prompt set", "category": "safety", "models": null}]}
{"id": "phi-4-paper", "type": "paper", "title": "Phi-4 Technical Report", "url": "https://arxiv.org/abs/2412.08905", "arxiv_id": "2412.08905", "models": ["microsoft/phi-4"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "simple_evals", "models": null}, {"name": "GPQA-Diamond", "category": "simple_evals", "models": null}, {"name": "MATH", "category": "simple_evals", "models": null}, {"name": "HumanEval", "category": "simple_evals", "models": null}, {"name": "MGSM", "category": "simple_evals", "models": null}, {"name": "SimpleQA", "category": "simple_evals", "models": null}, {"name": "DROP", "category": "simple_evals", "models": null}, {"name": "MMLU-Pro", "category": "internal_framework", "models": null}, {"name": "HumanEval+", "category": "internal_framework", "models": null}, {"name": "Arena-Hard", "category": "internal_framework", "models": null}, {"name": "LiveBench", "category": "internal_framework", "models": null}, {"name": "IFEval", "category": "internal_framework", "models": null}, {"name": "PhiBench", "category": "internal_proprietary", "models": null}, {"name": "TriviaQA", "category": "pretraining_ablation", "models": null}, {"name": "ARC-Challenge", "category": "pretraining_ablation", "models": null}, {"name": "MBPP", "category": "pretraining_ablation", "models": null}, {"name": "GSM8K", "category": "pretraining_ablation", "models": null}, {"name": "Internal Responsible AI eval (Grounding, Jailbreak DR1, multi-turn harm)", "category": "safety", "models": null}, {"name": "AMC-10/12 Nov 2024", "category": "math_probe", "models": null}]}
{"id": "phi-3-mini-card", "type": "model_card", "title": "Phi-3-mini-4k-instruct model card (report at aka.ms/phi3-tech-report, no arXiv ID)", "url": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", "arxiv_id": null, "models": ["microsoft/Phi-3-mini-4k-instruct"], "notes": null, "benchmarks": [{"name": "AGIEval", "category": "aggregate", "models": null}, {"name": "MMLU", "category": "aggregate", "models": null}, {"name": "BBH", "category": "aggregate", "models": null}, {"name": "ANLI", "category": "language_understanding", "models": null}, {"name": "HellaSwag", "category": "language_understanding", "models": null}, {"name": "ARC-Challenge", "category": "reasoning", "models": null}, {"name": "BoolQ", "category": "reasoning", "models": null}, {"name": "MedQA", "category": "reasoning", "models": null}, {"name": "OpenBookQA", "category": "reasoning", "models": null}, {"name": "PIQA", "category": "reasoning", "models": null}, {"name": "GPQA", "category": "reasoning", "models": null}, {"name": "SocialIQA", "category": "reasoning", "models": null}, {"name": "TruthfulQA", "category": "reasoning", "models": null}, {"name": "WinoGrande", "category": "reasoning", "models": null}, {"name": "TriviaQA", "category": "knowledge", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "Instruction Extra Hard", "category": "instruction_following", "models": null}, {"name": "Instruction Hard", "category": "instruction_following", "models": null}, {"name": "Instructions Challenge", "category": "instruction_following", "models": null}, {"name": "JSON Structure Output", "category": "instruction_following", "models": null}, {"name": "XML Structure Output", "category": "instruction_following", "models": null}]}
{"id": "jamba-paper", "type": "paper", "title": "Jamba: A Hybrid Transformer-Mamba Language Model", "url": "https://arxiv.org/abs/2403.19887", "arxiv_id": "2403.19887", "models": ["ai21labs/Jamba-v0.1"], "notes": "Base/pretrained checkpoint, not instruction-tuned; no safety/multilingual/alignment benchmarks.", "benchmarks": [{"name": "HellaSwag", "category": "commonsense", "models": null}, {"name": "WinoGrande", "category": "commonsense", "models": null}, {"name": "ARC-Easy", "category": "commonsense", "models": null}, {"name": "ARC-Challenge", "category": "commonsense", "models": null}, {"name": "PIQA", "category": "commonsense", "models": null}, {"name": "BoolQ", "category": "reading_comprehension", "models": null}, {"name": "QuAC", "category": "reading_comprehension", "models": null}, {"name": "Natural Questions", "category": "knowledge", "models": null}, {"name": "TruthfulQA", "category": "knowledge", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MMLU", "category": "aggregate", "models": null}, {"name": "BBH", "category": "aggregate", "models": null}, {"name": "Needle-in-a-haystack", "category": "long_context", "models": null}, {"name": "Trec-Fine", "category": "long_context", "models": null}, {"name": "NLU Intent", "category": "long_context", "models": null}, {"name": "Banking77", "category": "long_context", "models": null}, {"name": "CLINC150", "category": "long_context", "models": null}, {"name": "NarrativeQA", "category": "long_context", "models": null}, {"name": "LongFQA", "category": "long_context", "models": null}, {"name": "CUAD", "category": "long_context", "models": null}, {"name": "SFiction", "category": "long_context", "models": null}]}
{"id": "apertus-paper", "type": "paper", "title": "Apertus Technical Report", "url": "https://arxiv.org/abs/2509.14233", "arxiv_id": "2509.14233", "models": ["swiss-ai/Apertus-8B-2509"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "Global-MMLU", "category": "knowledge", "models": null}, {"name": "CulturalBench", "category": "knowledge", "models": null}, {"name": "SwitzerlandQA", "category": "knowledge", "models": null}, {"name": "BBH", "category": "reasoning", "models": null}, {"name": "DROP", "category": "reasoning", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MGSM", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "TruthfulQA", "category": "truthfulness", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "INCLUDE", "category": "multilingual", "models": null}, {"name": "FLORES+", "category": "multilingual", "models": null}, {"name": "ToxiGen", "category": "safety", "models": null}, {"name": "BBQ", "category": "safety", "models": null}, {"name": "Constitutional Harms Test", "category": "safety", "models": null}]}
{"id": "grok-1-blog", "type": "blog", "title": "\"Announcing Grok\" xAI blog (no formal paper; HF card is sparse)", "url": "https://x.ai/news/grok", "arxiv_id": null, "models": ["xai-org/grok-1"], "notes": "Benchmark table (GSM8K, MMLU, HumanEval, MATH) verified via user-provided screenshot of the original xAI announcement table (comparing Grok-0 33B, LLaMa 2 70B, Inflection-1, GPT-3.5, Grok-1, Palm 2, Claude 2, GPT-4). Very limited public benchmarking beyond this - no safety/multilingual/instruction-following disclosure.", "benchmarks": [{"name": "MMLU", "category": "knowledge_reasoning", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "2023 Hungarian national high school math final exam", "category": "out_of_distribution", "models": null}]}
{"id": "command-r-plus-card", "type": "model_card", "title": "Command R+ model card (Open LLM Leaderboard results)", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-plus", "arxiv_id": null, "models": ["CohereLabs/c4ai-command-r-plus"], "notes": "Card explicitly notes these don't capture RAG/multilingual/tool-use performance (pointed to a separate blog/Chatbot Arena instead, no numeric data given there).", "benchmarks": [{"name": "ARC-Challenge", "category": "open_llm_leaderboard", "models": null}, {"name": "HellaSwag", "category": "open_llm_leaderboard", "models": null}, {"name": "MMLU", "category": "open_llm_leaderboard", "models": null}, {"name": "TruthfulQA", "category": "open_llm_leaderboard", "models": null}, {"name": "WinoGrande", "category": "open_llm_leaderboard", "models": null}, {"name": "GSM8K", "category": "open_llm_leaderboard", "models": null}]}
{"id": "olmo-3-paper", "type": "paper", "title": "Olmo 3", "url": "https://arxiv.org/abs/2512.13961", "arxiv_id": "2512.13961", "models": ["allenai/Olmo-3-1025-7B"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "knowledge_reasoning", "models": null}, {"name": "MMLU-Pro", "category": "knowledge_reasoning", "models": null}, {"name": "BBH", "category": "knowledge_reasoning", "models": null}, {"name": "ARC-Challenge", "category": "knowledge_reasoning", "models": null}, {"name": "CommonsenseQA", "category": "knowledge_reasoning", "models": null}, {"name": "HellaSwag", "category": "knowledge_reasoning", "models": null}, {"name": "WinoGrande", "category": "knowledge_reasoning", "models": null}, {"name": "Lambada", "category": "knowledge_reasoning", "models": null}, {"name": "PIQA", "category": "knowledge_reasoning", "models": null}, {"name": "SocialIQA", "category": "knowledge_reasoning", "models": null}, {"name": "GPQA", "category": "knowledge_reasoning", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "AIME 2024", "category": "math", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "OMEGA", "category": "math", "models": null}, {"name": "DeepMind Math", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "HumanEval+", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "BigCodeBench", "category": "coding", "models": null}, {"name": "DS-1000", "category": "coding", "models": null}, {"name": "MultiPL-E", "category": "coding", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "DROP", "category": "qa_reading", "models": null}, {"name": "Natural Questions", "category": "qa_reading", "models": null}, {"name": "SQuAD", "category": "qa_reading", "models": null}, {"name": "CoQA", "category": "qa_reading", "models": null}, {"name": "MedMCQA", "category": "qa_reading", "models": null}, {"name": "MedQA", "category": "qa_reading", "models": null}, {"name": "SciQ", "category": "qa_reading", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "AlpacaEval 2.0", "category": "general", "models": null}, {"name": "Internal long-context eval suite", "category": "long_context", "models": null}]}
{"id": "hyperclovax-paper", "type": "paper", "title": "HyperCLOVA X 32B Think", "url": "https://arxiv.org/abs/2601.03286", "arxiv_id": "2601.03286", "models": ["naver-hyperclovax/HyperCLOVAX-SEED-Think-32B"], "notes": null, "benchmarks": [{"name": "KMMLU", "category": "korean", "models": null}, {"name": "KoBALT", "category": "korean", "models": null}, {"name": "CLIcK", "category": "korean", "models": null}, {"name": "HAERAE Bench", "category": "korean", "models": null}, {"name": "KCSAT 2026 (Math)", "category": "korean", "models": null}, {"name": "KoNET", "category": "korean", "models": null}, {"name": "FLORES+", "category": "korean", "models": null}, {"name": "MMLU", "category": "general", "models": null}, {"name": "HellaSwag", "category": "general", "models": null}, {"name": "PIQA", "category": "general", "models": null}, {"name": "K-MMBench", "category": "vision_language", "models": null}, {"name": "K-DTCBench", "category": "vision_language", "models": null}, {"name": "SEED-IMG", "category": "vision_language", "models": null}, {"name": "LLaVA-W", "category": "vision_language", "models": null}, {"name": "TextVQA", "category": "vision_language", "models": null}, {"name": "DocVQA", "category": "vision_language", "models": null}, {"name": "Video-MME", "category": "vision_language", "models": null}, {"name": "ChartVQA", "category": "vision_language", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "Terminal-Bench", "category": "agentic", "models": null}, {"name": "Terminal-Bench Hard", "category": "agentic", "models": null}]}
{"id": "gigachat-3.5-card", "type": "model_card", "title": "GigaChat3.5-432B-A28B model card (base + instruct benchmarks)", "url": "https://huggingface.co/ai-sage/GigaChat3.5-432B-A28B-base", "arxiv_id": null, "models": ["ai-sage/GigaChat3.5-432B-A28B-base", "ai-sage/GigaChat3.5-432B-A28B"], "notes": "Sber (Russian lab). Base and instruct checkpoints are evaluated on different suites; scoped per-model below.", "benchmarks": [{"name": "MMLU", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "MMLU-Pro", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "GPQA-Diamond", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "BBH", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "ARC-Challenge", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "ARC-Easy", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "HellaSwag", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "WinoGrande", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "DROP", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "TriviaQA", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "NQ-Open", "category": "base_general", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "Minerva Math", "category": "base_math", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "GSM8K", "category": "base_math", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "MGSM-ru", "category": "base_math", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "HumanEval", "category": "base_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "HumanEval+", "category": "base_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "MBPP", "category": "base_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "MBPP+", "category": "base_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "CRUXEval", "category": "base_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "LiveCodeBench", "category": "base_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B-base"]}, {"name": "TAU2-Bench", "category": "instruct_agentic", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "SWE-bench Verified", "category": "instruct_agentic", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "Terminal-Bench 2.0", "category": "instruct_agentic", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "LiveCodeBench v6", "category": "instruct_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "Natural Plan", "category": "instruct_coding", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "IFBench", "category": "instruct_general", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "MMLU-Pro", "category": "instruct_general", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "MATH-500", "category": "instruct_math", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "MERA Text", "category": "instruct_russian", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "Pollux", "category": "instruct_russian", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "RubQ_Ru", "category": "instruct_russian", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "Arena-Hard (Ru vs GPT-5)", "category": "instruct_russian", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}, {"name": "Ru LLM Arena", "category": "instruct_russian", "models": ["ai-sage/GigaChat3.5-432B-A28B"]}]}
{"id": "minicpm-papers", "type": "paper", "title": "MiniCPM4/4.1 report (2506.07900) + data-management companion paper (2602.09003) - no dedicated MiniCPM5 report exists", "url": "https://arxiv.org/abs/2506.07900", "arxiv_id": "2506.07900", "models": ["openbmb/MiniCPM5-1B"], "notes": "None of MiniCPM5-1B's 4 tagged arXiv IDs is actually titled 'MiniCPM5' - benchmarks are proxied from the MiniCPM4/4.1 report (predecessor generation, same architecture lineage/scale) plus a training-data companion paper.", "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMLU-Redux", "category": "knowledge", "models": null}, {"name": "CMMLU", "category": "knowledge", "models": null}, {"name": "C-Eval", "category": "knowledge", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "MATH-500", "category": "math", "models": null}, {"name": "AIME 2024", "category": "math", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "MultiPL-E", "category": "coding", "models": null}, {"name": "BBH", "category": "reasoning", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "RULER", "category": "long_context", "models": null}, {"name": "ARC-Challenge", "category": "commonsense", "models": null}, {"name": "ARC-Easy", "category": "commonsense", "models": null}, {"name": "CommonsenseQA", "category": "commonsense", "models": null}, {"name": "HellaSwag", "category": "commonsense", "models": null}, {"name": "OpenBookQA", "category": "commonsense", "models": null}, {"name": "PIQA", "category": "commonsense", "models": null}, {"name": "SocialIQA", "category": "commonsense", "models": null}, {"name": "WinoGrande", "category": "commonsense", "models": null}]}
{"id": "llama-nemotron-paper", "type": "paper", "title": "Llama-Nemotron: Efficient Reasoning Models", "url": "https://arxiv.org/abs/2505.00949", "arxiv_id": "2505.00949", "models": ["nvidia/Llama-3_3-Nemotron-Super-49B-v1"], "notes": "Distinct from the separate 'Nemotron 3' family already in this DB - this is NVIDIA's efficiency distillation of Meta's Llama 3.3.", "benchmarks": [{"name": "AIME 2024", "category": "math", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "MATH-500", "category": "math", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "BFCL v2", "category": "agentic", "models": null}, {"name": "IFEval", "category": "general", "models": null}, {"name": "Arena-Hard", "category": "general", "models": null}, {"name": "JudgeBench", "category": "llm_judge", "models": null}]}
{"id": "gemma-3n-card", "type": "model_card", "title": "Gemma 3n model card / blog (no dedicated arXiv report; same dataset-citation-tag style as Gemma 3)", "url": "https://ai.google.dev/gemma/docs/gemma-3n", "arxiv_id": null, "models": ["google/gemma-3n-E2B-it"], "notes": null, "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "Global-MMLU", "category": "knowledge", "models": null}, {"name": "Global-MMLU-Lite", "category": "knowledge", "models": null}, {"name": "TriviaQA", "category": "knowledge", "models": null}, {"name": "Natural Questions", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HellaSwag", "category": "reasoning_commonsense", "models": null}, {"name": "BoolQ", "category": "reasoning_commonsense", "models": null}, {"name": "PIQA", "category": "reasoning_commonsense", "models": null}, {"name": "SocialIQA", "category": "reasoning_commonsense", "models": null}, {"name": "WinoGrande", "category": "reasoning_commonsense", "models": null}, {"name": "ARC-Challenge", "category": "reasoning_commonsense", "models": null}, {"name": "ARC-Easy", "category": "reasoning_commonsense", "models": null}, {"name": "DROP", "category": "reasoning_commonsense", "models": null}, {"name": "BBH", "category": "reasoning_commonsense", "models": null}, {"name": "MGSM", "category": "math", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "HiddenMath", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MBPP", "category": "coding", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "Codegolf", "category": "coding", "models": null}, {"name": "MGSM", "category": "multilingual", "models": null}, {"name": "WMT24++", "category": "multilingual", "models": null}, {"name": "INCLUDE", "category": "multilingual", "models": null}, {"name": "Global-MMLU", "category": "multilingual", "models": null}, {"name": "Global-MMLU-Lite", "category": "multilingual", "models": null}, {"name": "ECLeKTic", "category": "multilingual", "models": null}]}
{"id": "grok-1.5-blog", "type": "blog", "title": "\"Grok-1.5\" xAI announcement (primary page 403-blocked to fetchers; sourced from secondary aggregators)", "url": "https://x.ai/news/grok-1.5", "arxiv_id": null, "models": ["grok-1.5"], "notes": "Benchmark table (MMLU, MATH, GSM8K, HumanEval) verified via user-provided screenshot of the original xAI announcement table (comparing Grok-1, Grok-1.5, Mistral Large, Claude 2, Claude 3 Sonnet, Gemini Pro 1.5, GPT-4, Claude 3 Opus).", "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "GSM8K", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}]}
{"id": "grok-2-blog", "type": "blog", "title": "\"Grok-2\" xAI announcement (primary page 403-blocked; sourced from secondary aggregators)", "url": "https://x.ai/news/grok-2", "arxiv_id": null, "models": ["grok-2"], "notes": "Benchmark table (GPQA, MMLU, MMLU-Pro, MATH, HumanEval, MMMU, MathVista, DocVQA) verified via user-provided screenshot of the original xAI announcement table (comparing Grok-1.5, Grok-2 mini, Grok-2, GPT-4 Turbo, Claude 3 Opus, Gemini Pro 1.5, Llama 3 405B, GPT-4o, Claude 3.5 Sonnet). Most granular scores in the table are for the Grok-2 mini variant rather than full Grok-2, but the benchmark names/columns are identical for both.", "benchmarks": [{"name": "MMLU", "category": "knowledge", "models": null}, {"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "GPQA", "category": "knowledge", "models": null}, {"name": "MATH", "category": "math", "models": null}, {"name": "HumanEval", "category": "coding", "models": null}, {"name": "MMMU", "category": "vision", "models": null}, {"name": "MathVista", "category": "vision", "models": null}, {"name": "DocVQA", "category": "vision", "models": null}]}
{"id": "grok-4-blog", "type": "blog", "title": "\"Grok 4\" xAI announcement (primary page 403-blocked; sourced from secondary aggregators)", "url": "https://x.ai/news/grok-4", "arxiv_id": null, "models": ["grok-4"], "notes": "GPQA, LiveCodeBench, USAMO 2025, HMMT 2025, AIME 2025, ARC-AGI-2, and HLE (incl. w/ Python+Internet tools variant) verified via user-provided screenshots of the original xAI announcement charts (these also show a 'Grok 4 Heavy w/ Python' multi-agent configuration scoring higher on every chart - treated as an eval setting of Grok 4, not a separate model). SWE-bench Verified, Vending-Bench, and Artificial Analysis Intelligence Index were not visible in the screenshots provided and remain sourced from secondary aggregators only.", "benchmarks": [{"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "HLE w/ Tools", "category": "knowledge", "models": null}, {"name": "ARC-AGI-2", "category": "abstraction_reasoning", "models": null}, {"name": "AIME 2025", "category": "math", "models": null}, {"name": "USAMO 2025", "category": "math", "models": null}, {"name": "LiveCodeBench", "category": "coding", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "Vending-Bench", "category": "agentic", "models": null}, {"name": "Artificial Analysis Intelligence Index", "category": "aggregate", "models": null}, {"name": "HMMT 2025", "category": "math", "models": null}]}
{"id": "grok-4.1-blog", "type": "blog", "title": "\"Grok 4.1\" xAI announcement (primary page 403-blocked; sourced from secondary aggregators)", "url": "https://x.ai/news/grok-4-1", "arxiv_id": null, "models": ["grok-4.1"], "notes": "EQ-Bench and Creative Writing v3 verified via user-provided screenshots of the original xAI announcement charts (both Elo-normalized leaderboards comparing Grok 4.1/4.1 Thinking against Kimi K2 Instruct, Gemini 2.5 Pro, GPT-5 Chat, Claude Opus 4/Sonnet 4.5, Grok 3/4, and others). Confirms the benchmark name is 'EQ-Bench', not 'EQ-Bench3' as earlier secondary sourcing suggested. LMArena was not visible in the screenshots provided and remains sourced from secondary aggregators only.", "benchmarks": [{"name": "LMArena", "category": "human_preference", "models": null}, {"name": "EQ-Bench", "category": "emotional_intelligence", "models": null}, {"name": "Creative Writing v3", "category": "creative_writing", "models": null}]}
{"id": "gemma-4-paper", "type": "paper", "title": "Gemma 4 Technical Report", "url": "https://arxiv.org/abs/2607.02770", "arxiv_id": "2607.02770", "models": ["google/gemma-4-31B-it", "google/gemma-4-26B-A4B-it", "google/gemma-4-12B-it", "google/gemma-4-E4B-it", "google/gemma-4-E2B-it"], "notes": "Supersedes the earlier gemma-4-family-card model-card source, which lacked per-model scoping (it applied audio benchmarks to all 5 sizes uniformly). Per the actual paper tables: LMArena (Table 4) is only reported for the 31B and 26B-A4B sizes; CoVoST/FLEURS audio benchmarks (Tables 7-8) are only reported for E2B, E4B, and 12B. Static/vision/long-context benchmarks (Tables 5, 6, 9) are reported for all 5 sizes.", "benchmarks": [{"name": "MMLU-Pro", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "BigBench Extra Hard", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "HLE w/ Tools", "category": "knowledge", "models": null}, {"name": "IFBench", "category": "knowledge", "models": null}, {"name": "IFEval", "category": "knowledge", "models": null}, {"name": "MMMLU", "category": "knowledge", "models": null}, {"name": "SciCode", "category": "knowledge", "models": null}, {"name": "AIME 2026", "category": "math", "models": null}, {"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "Codeforces", "category": "coding", "models": null}, {"name": "TAU2-Bench", "category": "agentic", "models": null}, {"name": "Terminal-Bench Hard", "category": "agentic", "models": null}, {"name": "MRCR v2", "category": "long_context", "models": null}, {"name": "RULER", "category": "long_context", "models": null}, {"name": "LOFT", "category": "long_context", "models": null}, {"name": "GraphWalks", "category": "long_context", "models": null}, {"name": "MTOB", "category": "long_context", "models": null}, {"name": "MMMU-Pro", "category": "vision", "models": null}, {"name": "MathVision", "category": "vision", "models": null}, {"name": "MedXpertQA-MM", "category": "vision", "models": null}, {"name": "InfographicVQA", "category": "vision", "models": null}, {"name": "OmniDocBench", "category": "vision", "models": null}, {"name": "LMArena", "category": "human_preference", "models": ["google/gemma-4-31B-it", "google/gemma-4-26B-A4B-it"]}, {"name": "CoVoST", "category": "audio", "models": ["google/gemma-4-E2B-it", "google/gemma-4-E4B-it", "google/gemma-4-12B-it"]}, {"name": "FLEURS", "category": "audio", "models": ["google/gemma-4-E2B-it", "google/gemma-4-E4B-it", "google/gemma-4-12B-it"]}]}
{"id": "muse-spark-1-1-eval-report", "type": "report", "title": "Muse Spark 1.1 Evaluation Report (Meta)", "url": "https://ai.meta.com/static-resource/muse-spark-1-1-evaluation-report", "arxiv_id": null, "models": ["muse-spark-1.1"], "notes": "Extracted from the primary evaluation report PDF (112 pp, high confidence); the announcement blog (https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/) has charts only. General capabilities from §5.2; safety/preparedness from Tables 1-2 (named external evals only; internal metrics like Curated CTFs, Internal Sycophancy, BioDesign Tools omitted). Name mappings: 'MCP Atlas'→MCP-Atlas, 'SWE-Bench Pro'→SWE-bench Pro, 'DeepSWE v1.1'→DeepSWE, 'VibeCodeBench v1.1'→VibeCodeBench, 'HealthBench Pro'→HealthBench Professional (assumed same OpenAI benchmark). Kept separate on purpose: ProtocolQA (report doesn't say Open-Ended) vs ProtocolQA Open-Ended; CharXiv Reasoning (explicit reasoning split) vs CharXiv; StrongREJECT v2 vs StrongReject (explicit v2); WebArena-Verified vs WebArena; Toolathlon-Verified vs Toolathlon; OSWorld 2.0 vs OSWorld-Verified.", "benchmarks": [{"name": "OSWorld-Verified", "category": "agentic_computer_use", "models": null}, {"name": "OSWorld 2.0", "category": "agentic_computer_use", "models": null}, {"name": "WebArena-Verified", "category": "agentic_computer_use", "models": null}, {"name": "GDPval-AA", "category": "agentic", "models": null}, {"name": "JobBench", "category": "agentic", "models": null}, {"name": "MCP-Atlas", "category": "agentic", "models": null}, {"name": "Toolathlon-Verified", "category": "agentic", "models": null}, {"name": "DeepSearchQA", "category": "agentic_search", "models": null}, {"name": "WideSearch", "category": "agentic_search", "models": null}, {"name": "Finance Agent v2", "category": "agentic_domain_specific", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding_agentic", "models": null}, {"name": "SWE-bench Pro", "category": "coding_agentic", "models": null}, {"name": "DeepSWE", "category": "coding_agentic", "models": null}, {"name": "VibeCodeBench", "category": "coding_agentic", "models": null}, {"name": "SWE-bench Verified Hard", "category": "coding_agentic", "models": null}, {"name": "HealthBench Professional", "category": "health", "models": null}, {"name": "CharXiv Reasoning", "category": "multimodal", "models": null}, {"name": "BabyVision", "category": "multimodal", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "MRCR v2", "category": "long_context", "models": null}, {"name": "MBCT", "category": "safety_bio", "models": null}, {"name": "VCT", "category": "safety_bio", "models": null}, {"name": "HPCT", "category": "safety_bio", "models": null}, {"name": "WMDP-Bio", "category": "safety_bio", "models": null}, {"name": "WMDP-Chem", "category": "safety_bio", "models": null}, {"name": "ProtocolQA", "category": "safety_bio", "models": null}, {"name": "SeqQA", "category": "safety_bio", "models": null}, {"name": "ABC Bench", "category": "safety_bio", "models": null}, {"name": "Cybench", "category": "safety_cyber", "models": null}, {"name": "CyberGym", "category": "safety_cyber", "models": null}, {"name": "ExploitGym", "category": "safety_cyber", "models": null}, {"name": "CyScenarioBench", "category": "safety_cyber", "models": null}, {"name": "AIRS-Bench", "category": "safety_alignment", "models": null}, {"name": "SHADE-Arena", "category": "safety_alignment", "models": null}, {"name": "GDM-Stealth", "category": "safety_alignment", "models": null}, {"name": "GDM Situational Awareness", "category": "safety_alignment", "models": null}, {"name": "MASK", "category": "safety_alignment", "models": null}, {"name": "Agentic Misalignment", "category": "safety_alignment", "models": null}, {"name": "StrongREJECT v2", "category": "safety_alignment", "models": null}, {"name": "FORTRESS", "category": "safety_alignment", "models": null}, {"name": "AgentHarm", "category": "safety_alignment", "models": null}, {"name": "AgentDojo", "category": "safety_alignment", "models": null}, {"name": "OR-Bench", "category": "safety_alignment", "models": null}, {"name": "SAVE-Bench", "category": "safety_alignment", "models": null}, {"name": "DeceptionBench", "category": "safety_alignment", "models": null}, {"name": "Alignment Faking", "category": "safety_alignment", "models": null}]}
{"id": "grok-4.5-blog", "type": "blog", "title": "\"Grok 4.5\" SpaceXAI announcement + Cursor co-announcement", "url": "https://x.ai/news/grok-4-5", "arxiv_id": null, "models": ["grok-4.5"], "notes": "Primary launch page 403-blocked to fetchers (standard for x.ai/news/*). Suite reconstructed from the Cursor co-announcement (https://cursor.com/blog/grok-4-5 — Cursor co-trained the model; names SWE-Bench Pro, Terminal-Bench, SWE-Bench multilingual, CursorBench, scores in charts only) plus secondary aggregators (MarkTechPost, Crypto Briefing) quoting the launch page: DeepSWE 1.0 62.0 / DeepSWE 1.1 53 (both logged as DeepSWE), Terminal-Bench 2.1 83.3, SWE-bench Pro 64.7, SWE Marathon 29.0 (#1), #1 on Harvey's Legal Agent Benchmark. Medium confidence — no screenshot of the primary chart yet; scores self-reported by xAI, no third-party verification at time of logging. Cursor footnote: CursorBench result inflated by accidental inclusion of an earlier Cursor codebase snapshot in training. Lab logged as 'xAI' for consistency (announcement branding is 'SpaceXAI').", "benchmarks": [{"name": "DeepSWE", "category": "coding_agentic", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding_agentic", "models": null}, {"name": "SWE-bench Pro", "category": "coding_agentic", "models": null}, {"name": "SWE-bench Multilingual", "category": "coding_agentic", "models": null}, {"name": "CursorBench", "category": "coding_agentic", "models": null}, {"name": "SWE Marathon", "category": "coding_agentic", "models": null}, {"name": "Legal Agent Benchmark", "category": "agentic_domain_specific", "models": null}]}
{"id": "lhtb-benchmark-report", "type": "report", "title": "Long-Horizon-Terminal-Bench (LHTB): Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Grading", "url": "https://zli12321.github.io/LHTB/", "arxiv_id": "2607.08964", "models": ["grok-4.5", "claude-sonnet-5", "claude-opus-4-8", "claude-fable-5", "gpt-5.6", "gpt-5.5", "MiniMaxAI/MiniMax-M3", "moonshotai/Kimi-K2.6", "zai-org/GLM-5.2", "zai-org/GLM-5.1", "deepseek-ai/DeepSeek-V4-Pro", "tencent/Hy3"], "notes": "Third-party benchmark report (Tencent HY LLM Frontier / IntelligenceLab, July 2026), not a lab release source: 46 containerized long-horizon terminal tasks, hidden replay-based verifiers, continuous partial-credit reward, Terminus-2 harness, 90-min budget. HF dataset: IntelligenceLab/Long-Horizon-Terminal-Bench. 'models' lists the evaluated models already present in this dataset (they keep their own release source_id); also evaluated but not dataset rows: Claude Sonnet 4.6, Kimi K2.7 Code, Qwen3.6 Plus, Qwen3.7 Max, Doubao Seed 2.1 Pro, Gemini 3.1 Pro, GPT-5.4, GPT-5.3 Codex, Grok 4.20. High confidence: primary sources (blog + leaderboard data + arXiv paper + HF dataset card). GPT-5.6 entry is the 'GPT-5.6-sol' variant.", "benchmarks": [{"name": "Long-Horizon Terminal-Bench", "category": "agentic_terminal", "models": null}]}
{"id": "edgebench-benchmark-report", "type": "report", "title": "EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments", "url": "https://edge-bench.org/", "arxiv_id": "2607.05155", "models": ["claude-opus-4-8", "gpt-5.5", "zai-org/GLM-5.1", "deepseek-ai/DeepSeek-V4-Pro"], "notes": "Third-party benchmark report (ByteDance Seed, July 2026), not a lab release source: 134 real-world long-horizon tasks (51 open-sourced) across 6 categories; agents iterate in executable environments for 12+ hours per task, scored over the full improvement trajectory (log-sigmoid time-scaling law). HF dataset: ByteDance-Seed/EdgeBench. 'models' lists the evaluated models already present in this dataset (they keep their own release source_id); also evaluated but not a dataset row: GPT-5.4. High confidence: primary sources (HF dataset card leaderboard + arXiv tech report).", "benchmarks": [{"name": "EdgeBench", "category": "agentic", "models": null}]}
{"id": "kimi-k3-report", "type": "report", "title": "Kimi K3 tech report / model card", "url": "https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf", "arxiv_id": null, "models": ["moonshotai/Kimi-K3"], "notes": "Benchmarks extracted from the HF model card evaluation table (full tech report PDF on GitHub; no arXiv id, tech blog at kimi.com/blog/kimi-k3). Card's 'HLE-Full' reports without/with tools -> logged as HLE + HLE w/ Tools. Card's tau3-Banking logged as TAU3-Bench Banking (banking split of TAU3-Bench); MCPMark-Verified logged as MCP-Mark Verified (verified split of existing MCP-Mark); 'APEX-Agents' collapsed to existing 'Apex-Agent' (same Mercor leaderboard). PostTrainBench is the full benchmark, distinct from existing PostTrainBench Lite. Kimi Code Bench 2.0 and PerceptionBench are in-house Moonshot benchmarks. High confidence: primary source.", "benchmarks": [{"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "CritPt", "category": "reasoning", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "HLE w/ Tools", "category": "knowledge", "models": null}, {"name": "DeepSWE", "category": "coding", "models": null}, {"name": "ProgramBench", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding", "models": null}, {"name": "FrontierSWE", "category": "coding", "models": null}, {"name": "SWE Marathon", "category": "coding", "models": null}, {"name": "PostTrainBench", "category": "coding", "models": null}, {"name": "MLS-Bench-Lite", "category": "coding", "models": null}, {"name": "SciCode", "category": "coding", "models": null}, {"name": "Kimi Code Bench 2.0", "category": "coding", "models": null}, {"name": "BrowseComp", "category": "agentic_search", "models": null}, {"name": "DeepSearchQA", "category": "agentic_search", "models": null}, {"name": "ResearchRubrics", "category": "agentic_search", "models": null}, {"name": "GDPval-AA v2", "category": "agentic", "models": null}, {"name": "Toolathlon-Verified", "category": "agentic", "models": null}, {"name": "MCP-Mark Verified", "category": "agentic", "models": null}, {"name": "MCP-Atlas", "category": "agentic", "models": null}, {"name": "AutomationBench", "category": "agentic", "models": null}, {"name": "JobBench", "category": "agentic", "models": null}, {"name": "AA-Briefcase", "category": "agentic", "models": null}, {"name": "Agent's Last Exam", "category": "agentic", "models": null}, {"name": "Apex-Agent", "category": "agentic", "models": null}, {"name": "OfficeQA Pro", "category": "agentic", "models": null}, {"name": "SpreadsheetBench 2", "category": "agentic", "models": null}, {"name": "OSWorld-Verified", "category": "agentic", "models": null}, {"name": "OSWorld 2.0", "category": "agentic", "models": null}, {"name": "SaaS-Bench", "category": "agentic", "models": null}, {"name": "TAU3-Bench Banking", "category": "agentic", "models": null}, {"name": "Harvey Lab-AA", "category": "agentic", "models": null}, {"name": "CorpFin v2", "category": "agentic", "models": null}, {"name": "Finance Agent v2", "category": "agentic", "models": null}, {"name": "Legal Research Bench", "category": "agentic", "models": null}, {"name": "WorldVQA ForceAnswer", "category": "vision", "models": null}, {"name": "OmniDocBench", "category": "vision", "models": null}, {"name": "PerceptionBench", "category": "vision", "models": null}, {"name": "Video-MME", "category": "video", "models": null}, {"name": "MMVU", "category": "video", "models": null}, {"name": "BabyVision", "category": "vision", "models": null}, {"name": "MMMU-Pro", "category": "vision", "models": null}, {"name": "CharXiv", "category": "vision", "models": null}, {"name": "MathVision", "category": "vision", "models": null}, {"name": "ZeroBench", "category": "vision", "models": null}]}
{"id": "inkling-model-card", "type": "model_card", "title": "Inkling model card (Thinking Machines Lab)", "url": "https://huggingface.co/thinkingmachines/Inkling", "arxiv_id": null, "models": ["thinkingmachines/Inkling"], "notes": "975B-total/41B-active multimodal MoE (text+image+audio in), Thinking Machines Lab's first open-weights model. Benchmarks from the card's evaluation table (results at effort=0.99). Card variants collapsed to canonical names: 'HLE (text only)'/'HLE (with tools)' -> HLE / HLE w/ Tools; 'SWEBench Pro (Public)' -> SWE-bench Pro; 'MMMU Pro (Standard 10)' -> MMMU-Pro; 'Charxiv RQ (with python)' -> CharXiv; 'BrowseComp (w/ Ctx)' -> BrowseComp; 'Tau 3 Banking' -> TAU3-Bench Banking; 'AA Omniscience' -> AA-Omniscience; 'GDPVal-AA v2' -> GDPval-AA v2. FORTRESS covers the card's Adversarial and Benign splits. High confidence: primary source.", "benchmarks": [{"name": "HLE", "category": "knowledge", "models": null}, {"name": "HLE w/ Tools", "category": "knowledge", "models": null}, {"name": "AIME 2026", "category": "math", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "SWE-bench Pro", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding", "models": null}, {"name": "GDPval-AA v2", "category": "agentic", "models": null}, {"name": "MCP-Atlas", "category": "agentic", "models": null}, {"name": "TAU3-Bench Banking", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic_search", "models": null}, {"name": "SimpleQA Verified", "category": "knowledge", "models": null}, {"name": "AA-Omniscience", "category": "knowledge", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "Global-MMLU-Lite", "category": "multilingual", "models": null}, {"name": "MMMU-Pro", "category": "vision", "models": null}, {"name": "CharXiv", "category": "vision", "models": null}, {"name": "Audio MC", "category": "audio", "models": null}, {"name": "MMAU", "category": "audio", "models": null}, {"name": "VoiceBench", "category": "audio", "models": null}, {"name": "FORTRESS", "category": "safety", "models": null}, {"name": "StrongReject", "category": "safety", "models": null}]}
{"id": "ax-k2-report", "type": "report", "title": "A.X K2 Technical Report", "url": "https://github.com/SKT-AI/A.X-K2/blob/main/A_X_K2_Tech_Report.pdf", "arxiv_id": null, "models": ["skt/A.X-K2"], "notes": "688B-A33B MoE, successor to A.X K1 (Korean Sovereign AI project). Benchmarks from the HF model card eval table (all in thinking mode; tech report PDF on GitHub, no arXiv). Card spellings canonicalized: 'AIME26' -> AIME 2026; 'Humanity's Last Exam' -> HLE; 'GPQA Diamond' -> GPQA-Diamond; 'LiveCodeBench v6 (Feb-May)' -> LiveCodeBench v6; 'BrowseComp (<=10 searches)' -> BrowseComp; 'tau2-Bench (Telecom)' -> TAU2-Bench Telecom (telecom split, AA-evaluated); 'KMO26 (1st round)' -> KMO 2026 (1st round). IMO 2025 (35/42, gold threshold) reported in card prose. High confidence: primary source.", "benchmarks": [{"name": "AIME 2026", "category": "math", "models": null}, {"name": "Apex", "category": "math", "models": null}, {"name": "Apex-shortlist", "category": "math", "models": null}, {"name": "KMO 2026 (1st round)", "category": "math", "models": null}, {"name": "IMO 2025", "category": "math", "models": null}, {"name": "KMMLU-Pro", "category": "knowledge", "models": null}, {"name": "KoBALT", "category": "knowledge", "models": null}, {"name": "CLIcK", "category": "knowledge", "models": null}, {"name": "HLE", "category": "knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "knowledge", "models": null}, {"name": "LiveCodeBench v6", "category": "coding", "models": null}, {"name": "SciCode", "category": "coding", "models": null}, {"name": "IFBench", "category": "general", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "RULER", "category": "long_context", "models": null}, {"name": "Needle-in-a-haystack", "category": "long_context", "models": null}, {"name": "TAU2-Bench Telecom", "category": "agentic", "models": null}, {"name": "BrowseComp", "category": "agentic", "models": null}]}
{"id": "minimax-h3-blog", "type": "blog", "title": "MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities", "url": "https://www.minimax.io/blog/minimax-h3", "arxiv_id": null, "models": ["MiniMaxAI/MiniMax-H3"], "notes": "Omni-modal audio+video GENERATION system (Hailuo family; text/image/video/audio in, 2K video with native stereo audio out) - not an LLM. Neither the HF model card nor the official blog reports any benchmark table, so no benchmark rows. MiniMax's launch communications cite third-party arena rankings only: #1 open model on LMArena Video Arena (T2V and I2V) and on Artificial Analysis video leaderboards (T2V Elo 1242, #2 overall; Video Editing Elo 1130, #1) as of 2026-07-31.", "benchmarks": []}
{"id": "muse-glimmer-30b-card", "type": "model_card", "title": "Muse Glimmer 30B model card", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "arxiv_id": null, "models": ["meta-models/Muse-Glimmer-30B"], "notes": "Full benchmark table extracted from the HF model card. Cyber-domain benchmarks (CyberGym, CyberBench) are not in the card's visible table but are described in the linked methodology report (https://research.meta.ai/static/muse-glimmer-methodology, 'Muse Glimmer per-benchmark details'); note the card itself only reports an 'inferred' (not directly measured) Cyber risk designation, so treat these two as lower-confidence. The two arXiv tags on the repo (2504.13181, 2602.06036) are not the model's own technical report -- they are the cited Perception Encoder architecture paper and DFlash speculative-decoding paper, component references rather than a Muse Glimmer paper.", "benchmarks": [{"name": "MCP-Atlas", "category": "agentic", "models": null}, {"name": "DeepSearchQA", "category": "agentic_search", "models": null}, {"name": "TAU3-Bench Banking", "category": "agentic", "models": null}, {"name": "WildClawBench", "category": "agentic", "models": null}, {"name": "GDPval-AA v2", "category": "agentic", "models": null}, {"name": "Gaia2", "category": "agentic", "models": null}, {"name": "SkillsBench", "category": "agentic", "models": null}, {"name": "OSWorld-Verified", "category": "agentic_computer_use", "models": null}, {"name": "SWE-bench Pro", "category": "coding_agentic", "models": null}, {"name": "SWE-bench Verified", "category": "coding_agentic", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding_agentic", "models": null}, {"name": "SciCode", "category": "coding_agentic", "models": null}, {"name": "CharXiv Reasoning", "category": "multimodal", "models": null}, {"name": "ScreenSpot Pro", "category": "multimodal", "models": null}, {"name": "OmniDocBench", "category": "multimodal", "models": null}, {"name": "MMMU-Pro", "category": "multimodal", "models": null}, {"name": "CIMemories", "category": "safety_alignment", "models": null}, {"name": "AgentDojo", "category": "safety_alignment", "models": null}, {"name": "MBCT", "category": "safety_bio", "models": null}, {"name": "HPCT", "category": "safety_bio", "models": null}, {"name": "VCT", "category": "safety_bio", "models": null}, {"name": "WMDP-Bio", "category": "safety_bio", "models": null}, {"name": "WMDP-Chem", "category": "safety_bio", "models": null}, {"name": "ProtocolQA", "category": "safety_bio", "models": null}, {"name": "CyberGym", "category": "safety_cyber", "models": null}, {"name": "CyberBench", "category": "safety_cyber", "models": null}, {"name": "IFBench", "category": "reasoning", "models": null}, {"name": "AIME 2026", "category": "reasoning", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning", "models": null}, {"name": "HLE", "category": "reasoning", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "BEAM-128K", "category": "long_context", "models": null}]}
{"id": "motif-3-technical-report", "type": "paper", "title": "Motif 3: Technical Report", "url": "https://arxiv.org/abs/2608.09119", "arxiv_id": "2608.09119", "models": ["Motif-Technologies/Motif-3"], "notes": "Benchmarks from Table 6 (main evaluation of the released Motif 3 post-trained/unified model). The report also includes a separate Table 4 pretraining evaluation (MMLU, MMLU-Pro, ARC-C, WinoGrande, HellaSwag, PIQA, GSM8K, MATH, HumanEval, MBPP) for the Motif-3-Base checkpoint, which is a distinct HF repo (Motif-Technologies/Motif-3-Base) not logged here since it wasn't the model requested. AA-Omniscience is reported here as two distinct sub-metrics (Accuracy and Non-Hallucination), unlike other sources in this dataset that log a single 'AA-Omniscience' score.", "benchmarks": [{"name": "GDPval-AA v2", "category": "agentic", "models": null}, {"name": "TAU2-Bench Telecom", "category": "agentic", "models": null}, {"name": "TAU3-Bench Banking", "category": "agentic", "models": null}, {"name": "ITBench-AA", "category": "agentic", "models": null}, {"name": "SWE-bench Verified", "category": "coding", "models": null}, {"name": "Terminal-Bench 2.1", "category": "coding", "models": null}, {"name": "SciCode", "category": "coding", "models": null}, {"name": "IMO-AnswerBench", "category": "reasoning_knowledge", "models": null}, {"name": "Apex-shortlist", "category": "reasoning_knowledge", "models": null}, {"name": "GPQA-Diamond", "category": "reasoning_knowledge", "models": null}, {"name": "HLE", "category": "reasoning_knowledge", "models": null}, {"name": "CritPt", "category": "reasoning_knowledge", "models": null}, {"name": "AA-Omniscience Accuracy", "category": "reasoning_knowledge", "models": null}, {"name": "AA-Omniscience Non-Hallucination", "category": "reasoning_knowledge", "models": null}, {"name": "AA-LCR", "category": "long_context", "models": null}, {"name": "IFBench", "category": "instruction_following", "models": null}]}