{
  "type": "benchpress-composition",
  "schema_version": 1,
  "name": "coding-mix",
  "created_at": "2026-07-07T18:09:56+00:00",
  "abilities": [
    "algorithmic_decomposition",
    "rule_constrained_reasoning"
  ],
  "sources": [
    {
      "benchmark": "HumanEval",
      "repo_id": "evalplus/humanevalplus",
      "config": "default",
      "split": "test",
      "revision": "d32357cf319e50e9c8d8dab5ea876c72b0fd321b",
      "gated": false,
      "license": "apache-2.0",
      "n_samples": 100
    },
    {
      "benchmark": "MBPP",
      "repo_id": "google-research-datasets/mbpp",
      "config": "full",
      "split": "test",
      "revision": "4bb6404fdc6cacfda99d4ac4205087b89d32030c",
      "gated": false,
      "license": "cc-by-4.0",
      "n_samples": 150
    },
    {
      "benchmark": "LiveCodeBench",
      "repo_id": "livecodebench/code_generation",
      "config": null,
      "split": "test",
      "revision": "bb83f1c3c643a772b7a2a8f07bd62c2aab76be8f",
      "gated": false,
      "license": "cc",
      "n_samples": 100
    }
  ],
  "combine": {
    "method": "concat",
    "seed": 42,
    "shuffle": true
  },
  "references": {
    "models": {
      "Claude-3.5-Sonnet": 0.7997,
      "Claude-Sonnet-4": 0.8314,
      "DeepSeek-v3": 0.604,
      "GPT-4o": 0.7206,
      "GPT-oss-20b": 0.8069,
      "Gemma-3-4B": 0.3671,
      "Llama-3.1-8B": 0.5489,
      "Phi-4-Mini": 0.5504,
      "Qwen2.5-72B": 0.784,
      "Qwen3-235B": 0.8086
    },
    "per_benchmark": {
      "HumanEval": {
        "DeepSeek-v3": 0.652,
        "GPT-4o": 0.902,
        "Qwen2.5-72B": 0.866,
        "Claude-3.5-Sonnet": 0.92,
        "Llama-3.1-8B": 0.665,
        "Phi-4-Mini": 0.744,
        "Gemma-3-4B": 0.36,
        "Claude-Sonnet-4": 0.963,
        "Qwen3-235B": 0.902,
        "GPT-oss-20b": 0.73,
        "GPT-oss-120b": 0.71,
        "KIMI-K2": 0.933
      },
      "MBPP": {
        "DeepSeek-v3": 0.754,
        "GPT-4o": 0.862,
        "Qwen2.5-72B": 0.882,
        "Qwen3-235B": 0.814,
        "Claude-3.5-Sonnet": 0.892,
        "Llama-3.1-8B": 0.694,
        "Phi-4-Mini": 0.653,
        "Gemma-3-4B": 0.46,
        "Claude-Sonnet-4": 0.858,
        "GPT-oss-20b": 0.878
      },
      "LiveCodeBench": {
        "DeepSeek-v3": 0.331,
        "GPT-4o": 0.327,
        "Qwen2.5-72B": 0.555,
        "Qwen3-235B": 0.707,
        "Claude-3.5-Sonnet": 0.541,
        "GPT-oss-120b": 0.878,
        "GPT-oss-20b": 0.777,
        "Llama-3.1-8B": 0.215,
        "Phi-4-Mini": 0.203,
        "Claude-Sonnet-4": 0.66,
        "KIMI-K2": 0.853,
        "GLM-4.7": 0.894,
        "Gemma-3-4B": 0.235
      }
    },
    "note": "models = n_samples-weighted mean; only models scored on every source are included."
  }
}