Auto-update results.csv

results.csv

@@ -11,6 +11,9 @@ SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/dataset
 SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro,OpenCode,[],True,https://github.com/anomalyco/opencode,Opus 4.8,Opus 4.8,False,1000,bf16,https://www.anthropic.com/news/claude-opus-4-8,harbor,https://github.com/harbor-framework/harbor,96.0,0.0,0.781,187217712.0,187209844.0,1280944.0,375708500.0,30686.0,39352.0,151.4104807499999,1950184.0,1950102.0,13343.0,3913630.0,1.58,409.0,319.0
 SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro,Claude Code,[],False,https://github.com/anthropics/claude-code,Sonnet 4.6,Sonnet 4.6,False,1000,bf16,https://www.anthropic.com/news/claude-sonnet-4-6,harbor,https://github.com/harbor-framework/harbor,96.0,1.0,0.5,190672390.0,184409111.0,1593112.0,376674613.0,40527.0,49734.0,184.42824125000004,1986170.0,1920928.0,16594.0,3923693.0,1.92,518.0,422.0
 SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro,Codex,[],True,https://github.com/openai/codex,GPT 5.5 - high,GPT 5.5 - high,False,9700,bf16,https://openai.com/index/introducing-gpt-5-5,harbor,https://github.com/harbor-framework/harbor,96.0,0.0,0.604,198924339.0,189578624.0,1560836.0,390063799.0,32914.0,39468.0,188.34296700000002,2072128.0,1974777.0,16258.0,4063164.0,1.96,411.0,342.0
+SWE-Bench Pro -- Ansible,scale-ai/swe-bench-pro,96,https://huggingface.co/datasets/scale-ai/swe-bench-pro,Claude Code,[],False,https://github.com/anthropics/claude-code,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,nvfp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,96.0,1.0,0.26,147452235.0,0.0,953241.0,148405476.0,33613.0,55241.0,5.84,1535960.0,0.0,9929.0,1545890.0,0.06,575.0,350.0
+SWE-Bench Pro -- Ansible,scale-ai/swe-bench-pro,96,https://huggingface.co/datasets/scale-ai/swe-bench-pro,OpenCode,[],True,https://github.com/anomalyco/opencode,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,nvfp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,96.0,0.0,0.385,140362798.0,0.0,1230697.0,141593495.0,23543.0,49289.0,6.54,1462112.0,0.0,12819.0,1474932.0,0.07,513.0,245.0
+SWE-Bench Pro -- Ansible,scale-ai/swe-bench-pro,96,https://huggingface.co/datasets/scale-ai/swe-bench-pro,Pi,[],True,https://github.com/earendil-works/pi/tree/main,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,nvfp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,96.0,0.0,0.333,104994429.0,0.0,1095179.0,106089608.0,15004.0,55176.0,4.17,1093691.0,0.0,11408.0,1105100.0,0.04,574.0,156.0
 SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro,Claude Code,[],False,https://github.com/anthropics/claude-code,Qwen3.6-35B-A3B-NVFP4,RedHatAI/Qwen3.6-35B-A3B-NVFP4,True,35,nvfp4,https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4,harbor,https://github.com/harbor-framework/harbor,96.0,6.0,0.458,367897697.0,0.0,1694885.0,369592582.0,39024.0,46758.0,9.64,3832267.0,0.0,17655.0,3849922.0,0.1,487.0,406.0
 SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro,OpenCode,[],True,https://github.com/anomalyco/opencode,Qwen3.6-35B-A3B-NVFP4,RedHatAI/Qwen3.6-35B-A3B-NVFP4,True,35,nvfp4,https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4,harbor,https://github.com/harbor-framework/harbor,96.0,4.0,0.375,207164679.0,0.0,1598703.0,208763382.0,49450.0,57287.0,12.21,2157965.0,0.0,16653.0,2174618.0,0.13,596.0,515.0
 SWE-Bench Pro -- Ansible,ScaleAI/SWE-bench_Pro,96,https://huggingface.co/datasets/ScaleAI/SWE-bench_Pro,OpenClaw,[],True,https://github.com/openclaw/openclaw,Qwen3.6-35B-A3B-NVFP4,RedHatAI/Qwen3.6-35B-A3B-NVFP4,True,35,nvfp4,https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4,harbor,https://github.com/harbor-framework/harbor,96.0,5.0,0.406,0.0,0.0,0.0,0.0,38085.0,50779.0,9.4,0.0,0.0,0.0,0.0,0.1,528.0,396.0
@@ -20,6 +23,9 @@ SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datas
 SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,OpenCode,[],True,https://github.com/anomalyco/opencode,Opus 4.8,Opus 4.8,False,1000,bf16,https://www.anthropic.com/news/claude-opus-4-8,harbor,https://github.com/harbor-framework/harbor,500.0,0.0,0.834,347751942.0,347725246.0,3640204.0,699117392.0,114633.0,171540.0,319.53304049999974,695503.0,695450.0,7280.0,1398234.0,0.64,343.0,229.0
 SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,Claude Code,[],False,https://github.com/anthropics/claude-code,Sonnet 4.6,Sonnet 4.6,False,1000,bf16,https://www.anthropic.com/news/claude-sonnet-4-6,harbor,https://github.com/harbor-framework/harbor,,,0.796,,,,,,,,,,,,,,
 SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,Codex,[],True,https://github.com/openai/codex,GPT 5.5 - high,GPT 5.5 - high,False,9700,bf16,https://openai.com/index/introducing-gpt-5-5,harbor,https://github.com/harbor-framework/harbor,500.0,0.0,0.798,415795756.0,386352384.0,3431224.0,805579364.0,92588.0,141637.0,443.32977200000016,831591.0,772704.0,6862.0,1611158.0,0.89,283.0,185.0
+SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,Claude Code,[],False,https://github.com/anthropics/claude-code,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,nvfp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,500.0,53.0,0.118,309857096.0,0.0,2416878.0,312273974.0,54338.0,555599.0,9.43,619714.0,0.0,4833.0,624547.0,0.02,1111.0,108.0
+SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,OpenCode,[],True,https://github.com/anomalyco/opencode,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,nvfp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,500.0,6.0,0.318,373713878.0,0.0,4051687.0,377765565.0,49103.0,543025.0,13.64,747427.0,0.0,8103.0,755531.0,0.03,1086.0,98.0
+SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,Pi,[],True,https://github.com/earendil-works/pi/tree/main,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,nvfp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,500.0,12.0,0.232,195246618.0,0.0,2726988.0,197973606.0,29236.0,458591.0,8.12,390493.0,0.0,5453.0,395947.0,0.02,917.0,58.0
 SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,Claude Code,[],False,https://github.com/anthropics/claude-code,Qwen3.6-35B-A3B-NVFP4,RedHatAI/Qwen3.6-35B-A3B-NVFP4,True,35,nvfp4,https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4,harbor,https://github.com/harbor-framework/harbor,500.0,1.0,0.632,1106618897.0,0.0,5733245.0,1112352142.0,122808.0,171897.0,34.11,2213237.0,0.0,11466.0,2224704.0,0.07,343.0,245.0
 SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,OpenClaw,[],True,https://github.com/openclaw/openclaw,Qwen3.6-35B-A3B-NVFP4,RedHatAI/Qwen3.6-35B-A3B-NVFP4,True,35,nvfp4,https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4,harbor,https://github.com/harbor-framework/harbor,500.0,3.0,0.588,0.0,0.0,0.0,0.0,120399.0,200354.0,33.44,0.0,0.0,0.0,0.0,0.07,400.0,240.0
 SWE-Bench Verified,SWE-bench/SWE-bench_Verified,500,https://huggingface.co/datasets/SWE-bench/SWE-bench_Verified,OpenCode,[],True,https://github.com/anomalyco/opencode,Qwen3.6-35B-A3B-NVFP4,RedHatAI/Qwen3.6-35B-A3B-NVFP4,True,35,nvfp4,https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4,harbor,https://github.com/harbor-framework/harbor,500.0,4.0,0.548,469806650.0,0.0,4937761.0,474744411.0,120473.0,185168.0,29.75,939613.0,0.0,9875.0,949488.0,0.06,370.0,240.0