{
  "schema_version": 2,
  "model": {
    "label": "Qwen 3.5 · INT4 · D+",
    "base_model": "Qwen3.5-9B",
    "quantization": "Q4_K_M",
    "parameter_billions": 9,
    "text_weights_bytes": 5680522464,
    "text_weights_gb": 5.680522464,
    "text_weights_gib": 5.290398806333542,
    "size_note": "仅语言模型 GGUF 文件；不包括运行时显存、KV 缓存、视觉模块或应用资源。",
    "display_spec": "量化 9B · INT4 · 5.68 GB"
  },
  "historical": {
    "release": "2024-11-25",
    "question_count": 232,
    "question_mean": {
      "bare": 71.05336719632224,
      "dplus": 85.46142246128522,
      "delta": 14.408055264962982
    },
    "task_mean": {
      "bare": 62.526677680988485,
      "dplus": 80.56527787979786,
      "delta": 18.038600198809377
    },
    "full_credit": {
      "bare": 160,
      "dplus": 181,
      "delta": 21
    },
    "note": "同题集历史开发记录对比。无最终答案按 0 分计；D+ 结果为多轮版本迭代汇总。该变化包含作答完成度与得分的变化，不是冻结配置的独立消融实验。"
  },
  "leaderboard": {
    "source": {
      "repository": "LiveBench/livebench.github.io",
      "path": "public/table_2024_11_25.csv",
      "url": "https://github.com/LiveBench/livebench.github.io/blob/main/public/table_2024_11_25.csv",
      "blob_sha": "c2ab984a126c0c75b3b6eb527746cccd4a4d2291",
      "retrieved_date": "2026-11-12",
      "release": "2024-11-25",
      "source_kind": "Official historical LiveBench source CSV",
      "row_count": 65
    },
    "metric": "Math：AMPS_Hard、math_comp、olympiad 三任务等权平均",
    "shafu_score": 80.56527787979786,
    "reference_count": 65,
    "below_count": 63,
    "explicit_larger_count": 14,
    "size_count_rule": "至少22项：14项模型标识明确标注大于9B的配置，加上官方已核实671B的DeepSeek V3和V3-0324，以及Phi-4、Mistral Large 2411、Mistral Small 2501/2503、Command R及Command R+六项官方规格。R1分数较高不计入；Sonar仅确认基座、未锁定历史接口，不计入。",
    "caveat": "历史数学子项数值参考，SHAFU 为多轮开发且使用工具，运行条件与公开配置不同；不是官方排名、当前总榜或同条件对战。",
    "rows": [
      {
        "model_id": "gemini-2.5-pro-exp-03-25",
        "display_label": "Gemini 2.5 Pro Exp 03-25",
        "score_percent": 90.19833333333334,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 87,
          "math_comp": 95.833,
          "olympiad": 87.762
        },
        "shafu_difference_points": -9.633055453535476,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "deepseek-r1",
        "display_label": "DeepSeek R1",
        "score_percent": 80.71300000000001,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 88,
          "math_comp": 88.542,
          "olympiad": 65.597
        },
        "shafu_difference_points": -0.1477221202021468,
        "parameters_b_total": 671,
        "activated_parameters_b": 37,
        "architecture": "MoE",
        "weights_open": true,
        "parameters_source": "https://huggingface.co/deepseek-ai/DeepSeek-R1",
        "size_label": "671B · MoE",
        "parameter_status": "vendor_confirmed"
      },
      {
        "model_id": "o1-2024-12-17-high",
        "display_label": "o1 (High)",
        "score_percent": 80.32266666666668,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 87,
          "math_comp": 89.583,
          "olympiad": 64.385
        },
        "shafu_difference_points": 0.24261121313118394,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "claude-3-7-sonnet-thinking",
        "display_label": "Claude 3.7 Sonnet Thinking",
        "score_percent": 78.99900000000001,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 84,
          "math_comp": 81.25,
          "olympiad": 71.747
        },
        "shafu_difference_points": 1.5662778797978518,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "qwq-32b",
        "display_label": "QwQ 32B",
        "score_percent": 77.82000000000001,
        "parameters_b_from_identifier": 32,
        "tasks": {
          "AMPS_Hard": 87,
          "math_comp": 91.667,
          "olympiad": 54.793
        },
        "shafu_difference_points": 2.745277879797854,
        "size_label": "32B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "o3-mini-2025-01-31-high",
        "display_label": "o3-mini (High)",
        "score_percent": 77.289,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 88,
          "math_comp": 91.667,
          "olympiad": 52.2
        },
        "shafu_difference_points": 3.27627787979786,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gemini-2.0-flash-thinking-exp-01-21",
        "display_label": "Gemini 2.0 Flash Thinking",
        "score_percent": 75.85233333333333,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 83,
          "math_comp": 85.417,
          "olympiad": 59.14
        },
        "shafu_difference_points": 4.712944546464527,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "deepseek-v3-0324",
        "display_label": "DeepSeek V3 0324",
        "score_percent": 73.52,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 82,
          "math_comp": 81.25,
          "olympiad": 57.31
        },
        "shafu_difference_points": 7.045277879797865,
        "parameters_b_total": 671,
        "activated_parameters_b": 37,
        "architecture": "MoE",
        "weights_open": true,
        "parameters_source": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324",
        "size_label": "671B · MoE",
        "parameter_status": "vendor_confirmed"
      },
      {
        "model_id": "o3-mini-2025-01-31-medium",
        "display_label": "o3-mini-2025-01-31-medium",
        "score_percent": 72.37,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 86,
          "math_comp": 88.542,
          "olympiad": 42.568
        },
        "shafu_difference_points": 8.195277879797857,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gemini-exp-1206",
        "display_label": "gemini-exp-1206",
        "score_percent": 72.358,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 81,
          "math_comp": 69.792,
          "olympiad": 66.282
        },
        "shafu_difference_points": 8.207277879797857,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gemini-2.0-pro-exp-02-05",
        "display_label": "gemini-2.0-pro-exp-02-05",
        "score_percent": 70.96633333333334,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 80,
          "math_comp": 69.792,
          "olympiad": 63.107
        },
        "shafu_difference_points": 9.598944546464523,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gpt-4.5-preview",
        "display_label": "GPT-4.5 Preview",
        "score_percent": 69.329,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 81,
          "math_comp": 70.833,
          "olympiad": 56.154
        },
        "shafu_difference_points": 11.236277879797868,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gemini-2.0-flash",
        "display_label": "gemini-2.0-flash",
        "score_percent": 65.61933333333333,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 84,
          "math_comp": 69.792,
          "olympiad": 43.066
        },
        "shafu_difference_points": 14.945944546464531,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "claude-3-7-sonnet",
        "display_label": "claude-3-7-sonnet",
        "score_percent": 63.26466666666666,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 68,
          "math_comp": 53.125,
          "olympiad": 68.669
        },
        "shafu_difference_points": 17.300611213131198,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "o3-mini-2025-01-31-low",
        "display_label": "o3-mini-2025-01-31-low",
        "score_percent": 63.06366666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 86,
          "math_comp": 78.125,
          "olympiad": 25.066
        },
        "shafu_difference_points": 17.50161121313119,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "o1-mini-2024-09-12",
        "display_label": "o1-mini-2024-09-12",
        "score_percent": 61.992333333333335,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 81,
          "math_comp": 77.083,
          "olympiad": 27.894
        },
        "shafu_difference_points": 18.572944546464527,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "hunyuan-turbos-20250313",
        "display_label": "hunyuan-turbos-20250313",
        "score_percent": 61.980333333333334,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 75,
          "math_comp": 68.75,
          "olympiad": 42.191
        },
        "shafu_difference_points": 18.584944546464527,
        "size_label": "该版本未核实",
        "parameter_status": "historical_version_unverified",
        "size_note": "不以系列或其他版本规格代替"
      },
      {
        "model_id": "deepseek-v3",
        "display_label": "deepseek-v3",
        "score_percent": 60.54,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 67,
          "math_comp": 60.417,
          "olympiad": 54.203
        },
        "shafu_difference_points": 20.025277879797862,
        "parameters_b_total": 671,
        "activated_parameters_b": 37,
        "architecture": "MoE",
        "weights_open": true,
        "parameters_source": "https://huggingface.co/deepseek-ai/DeepSeek-V3",
        "size_label": "671B · MoE",
        "parameter_status": "vendor_confirmed"
      },
      {
        "model_id": "gemini-2.0-flash-exp",
        "display_label": "gemini-2.0-flash-exp",
        "score_percent": 60.391,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 75,
          "math_comp": 64.583,
          "olympiad": 41.59
        },
        "shafu_difference_points": 20.174277879797863,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "chatgpt-4o-latest-2025-03-27",
        "display_label": "chatgpt-4o-latest-2025-03-27",
        "score_percent": 59.52333333333333,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 74,
          "math_comp": 67.708,
          "olympiad": 36.862
        },
        "shafu_difference_points": 21.041944546464528,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "deepseek-r1-distill-qwen-32b",
        "display_label": "DeepSeek R1 Distill Qwen 32B",
        "score_percent": 59.442,
        "parameters_b_from_identifier": 32,
        "tasks": {
          "AMPS_Hard": 68,
          "math_comp": 71.875,
          "olympiad": 38.451
        },
        "shafu_difference_points": 21.12327787979786,
        "size_label": "32B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "qwen2.5-max",
        "display_label": "qwen2.5-max",
        "score_percent": 58.34966666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 66,
          "math_comp": 55.208,
          "olympiad": 53.841
        },
        "shafu_difference_points": 22.21561121313119,
        "size_label": "官方未公开",
        "parameter_status": "not_public",
        "size_note": "MoE · 总参数量未公开",
        "architecture": "MoE",
        "parameters_source": "https://qwenlm.github.io/blog/qwen2.5-max/"
      },
      {
        "model_id": "qwq-32b-preview",
        "display_label": "qwq-32b-preview",
        "score_percent": 58.26499999999999,
        "parameters_b_from_identifier": 32,
        "tasks": {
          "AMPS_Hard": 75.5,
          "math_comp": 55.729,
          "olympiad": 43.566
        },
        "shafu_difference_points": 22.300277879797868,
        "size_label": "32B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "deepseek-r1-distill-llama-70b",
        "display_label": "DeepSeek R1 Distill Llama 70B",
        "score_percent": 58.10766666666667,
        "parameters_b_from_identifier": 70,
        "tasks": {
          "AMPS_Hard": 65,
          "math_comp": 78.125,
          "olympiad": 31.198
        },
        "shafu_difference_points": 22.457611213131194,
        "size_label": "70B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "gemini-2.0-flash-lite",
        "display_label": "gemini-2.0-flash-lite",
        "score_percent": 58.09166666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 70,
          "math_comp": 63.542,
          "olympiad": 40.733
        },
        "shafu_difference_points": 22.473611213131193,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "learnlm-1.5-pro-experimental",
        "display_label": "learnlm-1.5-pro-experimental",
        "score_percent": 57.75,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 67,
          "math_comp": 50,
          "olympiad": 56.25
        },
        "shafu_difference_points": 22.81527787979786,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gemini-2.0-flash-lite-preview-02-05",
        "display_label": "gemini-2.0-flash-lite-preview-02-05",
        "score_percent": 55.543,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 76,
          "math_comp": 61.458,
          "olympiad": 29.171
        },
        "shafu_difference_points": 25.022277879797862,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "perplexity-sonar-pro",
        "display_label": "perplexity-sonar-pro",
        "score_percent": 55.437333333333335,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 49,
          "math_comp": 56.25,
          "olympiad": 61.062
        },
        "shafu_difference_points": 25.127944546464526,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gemma-3-27b-it",
        "display_label": "Gemma 3 27B IT",
        "score_percent": 55.391999999999996,
        "parameters_b_from_identifier": 27,
        "tasks": {
          "AMPS_Hard": 74,
          "math_comp": 64.583,
          "olympiad": 27.593
        },
        "shafu_difference_points": 25.173277879797865,
        "size_label": "27B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "grok-2-1212",
        "display_label": "grok-2-1212",
        "score_percent": 54.879,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 67,
          "math_comp": 46.875,
          "olympiad": 50.762
        },
        "shafu_difference_points": 25.686277879797863,
        "size_label": "该版本未核实",
        "parameter_status": "historical_version_unverified",
        "size_note": "不以系列或其他版本规格代替"
      },
      {
        "model_id": "dracarys2-72b-instruct",
        "display_label": "dracarys2-72b-instruct",
        "score_percent": 54.665,
        "parameters_b_from_identifier": 72,
        "tasks": {
          "AMPS_Hard": 67,
          "math_comp": 53.125,
          "olympiad": 43.87
        },
        "shafu_difference_points": 25.900277879797862,
        "size_label": "72B",
        "parameter_status": "model_identifier",
        "size_note": "官方模型卡确认基座规模",
        "parameters_source": "https://huggingface.co/abacusai/Dracarys2-72B-Instruct"
      },
      {
        "model_id": "qwen2.5-72b-instruct-turbo",
        "display_label": "Qwen 2.5 72B Instruct Turbo",
        "score_percent": 54.29266666666666,
        "parameters_b_from_identifier": 72,
        "tasks": {
          "AMPS_Hard": 64,
          "math_comp": 56.25,
          "olympiad": 42.628
        },
        "shafu_difference_points": 26.2726112131312,
        "size_label": "72B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "claude-3-5-sonnet-20241022",
        "display_label": "claude-3-5-sonnet-20241022",
        "score_percent": 52.279333333333334,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 53,
          "math_comp": 45.833,
          "olympiad": 58.005
        },
        "shafu_difference_points": 28.285944546464528,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gpt-4o-2024-08-06",
        "display_label": "gpt-4o-2024-08-06",
        "score_percent": 49.544000000000004,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 57,
          "math_comp": 50,
          "olympiad": 41.632
        },
        "shafu_difference_points": 31.021277879797857,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "step-2-16k-202411",
        "display_label": "step-2-16k-202411",
        "score_percent": 48.77266666666666,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 53.333,
          "math_comp": 50.685,
          "olympiad": 42.3
        },
        "shafu_difference_points": 31.792611213131202,
        "size_label": "该版本未核实",
        "parameter_status": "historical_version_unverified",
        "size_note": "不以系列或其他版本规格代替"
      },
      {
        "model_id": "chatgpt-4o-latest-2025-01-29",
        "display_label": "chatgpt-4o-latest-2025-01-29",
        "score_percent": 48.01666666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 54,
          "math_comp": 44.792,
          "olympiad": 45.258
        },
        "shafu_difference_points": 32.54861121313119,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "qwen2.5-coder-32b-instruct",
        "display_label": "Qwen 2.5 Coder 32B Instruct",
        "score_percent": 46.61433333333334,
        "parameters_b_from_identifier": 32,
        "tasks": {
          "AMPS_Hard": 47,
          "math_comp": 51.042,
          "olympiad": 41.801
        },
        "shafu_difference_points": 33.95094454646452,
        "size_label": "32B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "grok-beta",
        "display_label": "grok-beta",
        "score_percent": 45.84233333333333,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 57,
          "math_comp": 37.5,
          "olympiad": 43.027
        },
        "shafu_difference_points": 34.72294454646453,
        "size_label": "该版本未核实",
        "parameter_status": "historical_version_unverified",
        "size_note": "不以系列或其他版本规格代替"
      },
      {
        "model_id": "claude-3-opus-20240229",
        "display_label": "claude-3-opus-20240229",
        "score_percent": 43.61933333333334,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 42,
          "math_comp": 37.5,
          "olympiad": 51.358
        },
        "shafu_difference_points": 36.945944546464524,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gpt-4o-2024-11-20",
        "display_label": "gpt-4o-2024-11-20",
        "score_percent": 42.86666666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 50,
          "math_comp": 46.875,
          "olympiad": 31.725
        },
        "shafu_difference_points": 37.698611213131194,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "mistral-large-2411",
        "display_label": "mistral-large-2411",
        "score_percent": 42.54966666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 43,
          "math_comp": 32.292,
          "olympiad": 52.357
        },
        "shafu_difference_points": 38.015611213131194,
        "parameters_b_total": 123,
        "parameters_source": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "123B"
      },
      {
        "model_id": "llama-3.3-70b-instruct-turbo",
        "display_label": "Llama 3.3 70B Instruct Turbo",
        "score_percent": 42.23666666666667,
        "parameters_b_from_identifier": 70,
        "tasks": {
          "AMPS_Hard": 53,
          "math_comp": 41.667,
          "olympiad": 32.043
        },
        "shafu_difference_points": 38.32861121313119,
        "size_label": "70B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "phi-4",
        "display_label": "Phi-4",
        "score_percent": 41.98466666666666,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 57,
          "math_comp": 42.708,
          "olympiad": 26.246
        },
        "shafu_difference_points": 38.5806112131312,
        "parameters_b_total": 14,
        "parameters_source": "https://huggingface.co/microsoft/phi-4",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "14B"
      },
      {
        "model_id": "perplexity-sonar",
        "display_label": "perplexity-sonar",
        "score_percent": 41.626,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 38,
          "math_comp": 55.479,
          "olympiad": 31.399
        },
        "shafu_difference_points": 38.939277879797864,
        "size_label": "70B 基座",
        "parameter_status": "base_only_historical_version_unverified",
        "size_note": "官方 2025 介绍；历史接口未锁定",
        "base_parameters_b": 70,
        "parameters_source": "https://www.perplexity.ai/en-GB/hub/blog/meet-new-sonar"
      },
      {
        "model_id": "meta-llama-3.1-405b-instruct-turbo",
        "display_label": "Llama 3.1 405B Instruct Turbo",
        "score_percent": 41.071333333333335,
        "parameters_b_from_identifier": 405,
        "tasks": {
          "AMPS_Hard": 49,
          "math_comp": 42.708,
          "olympiad": 31.506
        },
        "shafu_difference_points": 39.493944546464526,
        "size_label": "405B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "dracarys2-llama-3.1-70b-instruct",
        "display_label": "dracarys2-llama-3.1-70b-instruct",
        "score_percent": 40.29933333333333,
        "parameters_b_from_identifier": 70,
        "tasks": {
          "AMPS_Hard": 52,
          "math_comp": 33.333,
          "olympiad": 35.565
        },
        "shafu_difference_points": 40.26594454646453,
        "size_label": "70B",
        "parameter_status": "model_identifier",
        "size_note": "官方模型卡确认基座规模",
        "parameters_source": "https://huggingface.co/abacusai/Dracarys2-Llama-3.1-70B-Instruct"
      },
      {
        "model_id": "mistral-small-2501",
        "display_label": "mistral-small-2501",
        "score_percent": 39.88666666666666,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 45,
          "math_comp": 36.458,
          "olympiad": 38.202
        },
        "shafu_difference_points": 40.6786112131312,
        "parameters_b_total": 24,
        "parameters_source": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "24B"
      },
      {
        "model_id": "qwen2.5-7b-instruct-turbo",
        "display_label": "qwen2.5-7b-instruct-turbo",
        "score_percent": 39.511,
        "parameters_b_from_identifier": 7,
        "tasks": {
          "AMPS_Hard": 52.667,
          "math_comp": 42.466,
          "olympiad": 23.4
        },
        "shafu_difference_points": 41.05427787979786,
        "size_label": "7B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "mistral-small-2503",
        "display_label": "mistral-small-2503",
        "score_percent": 39.434,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 38,
          "math_comp": 40.625,
          "olympiad": 39.677
        },
        "shafu_difference_points": 41.131277879797864,
        "parameters_b_total": 24,
        "parameters_source": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "24B"
      },
      {
        "model_id": "amazon.nova-pro-v1:0",
        "display_label": "amazon.nova-pro-v1:0",
        "score_percent": 38.04333333333333,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 45,
          "math_comp": 40.625,
          "olympiad": 28.505
        },
        "shafu_difference_points": 42.52194454646453,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "amazon.nova-lite-v1:0",
        "display_label": "amazon.nova-lite-v1:0",
        "score_percent": 36.699333333333335,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 50,
          "math_comp": 35.417,
          "olympiad": 24.681
        },
        "shafu_difference_points": 43.865944546464526,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "gpt-4o-mini-2024-07-18",
        "display_label": "gpt-4o-mini-2024-07-18",
        "score_percent": 36.31066666666667,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 55,
          "math_comp": 31.25,
          "olympiad": 22.682
        },
        "shafu_difference_points": 44.25461121313119,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "claude-3-5-haiku-20241022",
        "display_label": "claude-3-5-haiku-20241022",
        "score_percent": 35.535000000000004,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 42,
          "math_comp": 40.625,
          "olympiad": 23.98
        },
        "shafu_difference_points": 45.03027787979786,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "meta-llama-3.1-70b-instruct-turbo",
        "display_label": "meta-llama-3.1-70b-instruct-turbo",
        "score_percent": 34.72,
        "parameters_b_from_identifier": 70,
        "tasks": {
          "AMPS_Hard": 41,
          "math_comp": 35.417,
          "olympiad": 27.743
        },
        "shafu_difference_points": 45.84527787979786,
        "size_label": "70B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "amazon.nova-micro-v1:0",
        "display_label": "amazon.nova-micro-v1:0",
        "score_percent": 34.49433333333334,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 48,
          "math_comp": 33.333,
          "olympiad": 22.15
        },
        "shafu_difference_points": 46.070944546464524,
        "size_label": "官方未公开",
        "parameter_status": "not_public"
      },
      {
        "model_id": "azerogpt",
        "display_label": "azerogpt",
        "score_percent": 32.85,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 38,
          "math_comp": 33.333,
          "olympiad": 27.217
        },
        "shafu_difference_points": 47.71527787979786,
        "size_label": "官方未公开",
        "parameter_status": "not_public",
        "parameters_source": "https://azero.soundai.com/prod_azerogpt.html"
      },
      {
        "model_id": "gemma-2-27b-it",
        "display_label": "Gemma 2 27B IT",
        "score_percent": 26.457333333333334,
        "parameters_b_from_identifier": 27,
        "tasks": {
          "AMPS_Hard": 34,
          "math_comp": 18.75,
          "olympiad": 26.622
        },
        "shafu_difference_points": 54.10794454646452,
        "size_label": "27B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "command-r-plus-08-2024",
        "display_label": "command-r-plus-08-2024",
        "score_percent": 21.272000000000002,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 21,
          "math_comp": 20.833,
          "olympiad": 21.983
        },
        "shafu_difference_points": 59.293277879797856,
        "parameters_b_total": 104,
        "parameters_source": "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "104B"
      },
      {
        "model_id": "gemma-2-9b-it",
        "display_label": "gemma-2-9b-it",
        "score_percent": 19.804333333333336,
        "parameters_b_from_identifier": 9,
        "tasks": {
          "AMPS_Hard": 30,
          "math_comp": 13.542,
          "olympiad": 15.871
        },
        "shafu_difference_points": 60.76094454646453,
        "size_label": "9B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "command-r-08-2024",
        "display_label": "command-r-08-2024",
        "score_percent": 19.392333333333333,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 15,
          "math_comp": 18.75,
          "olympiad": 24.427
        },
        "shafu_difference_points": 61.17294454646453,
        "parameters_b_total": 32,
        "parameters_source": "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "32B"
      },
      {
        "model_id": "meta-llama-3.1-8b-instruct-turbo",
        "display_label": "meta-llama-3.1-8b-instruct-turbo",
        "score_percent": 18.30666666666667,
        "parameters_b_from_identifier": 8,
        "tasks": {
          "AMPS_Hard": 21,
          "math_comp": 19.792,
          "olympiad": 14.128
        },
        "shafu_difference_points": 62.25861121313119,
        "size_label": "8B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      },
      {
        "model_id": "phi-3-small-8k-instruct",
        "display_label": "phi-3-small-8k-instruct",
        "score_percent": 17.580333333333332,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 18,
          "math_comp": 16.667,
          "olympiad": 18.074
        },
        "shafu_difference_points": 62.98494454646453,
        "parameters_b_total": 7,
        "parameters_source": "https://huggingface.co/microsoft/Phi-3-small-8k-instruct",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "7B"
      },
      {
        "model_id": "phi-3-mini-128k-instruct",
        "display_label": "phi-3-mini-128k-instruct",
        "score_percent": 15.718666666666666,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 18,
          "math_comp": 12.5,
          "olympiad": 16.656
        },
        "shafu_difference_points": 64.8466112131312,
        "parameters_b_total": 3.8,
        "parameters_source": "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/blob/main/README.md",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "3.8B"
      },
      {
        "model_id": "phi-3-mini-4k-instruct",
        "display_label": "phi-3-mini-4k-instruct",
        "score_percent": 14.957999999999998,
        "parameters_b_from_identifier": null,
        "tasks": {
          "AMPS_Hard": 19,
          "math_comp": 14.583,
          "olympiad": 11.291
        },
        "shafu_difference_points": 65.60727787979786,
        "parameters_b_total": 3.8,
        "parameters_source": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct",
        "parameter_status": "vendor_confirmed",
        "weights_open": true,
        "size_label": "3.8B"
      },
      {
        "model_id": "olmo-2-1124-13b-instruct",
        "display_label": "olmo-2-1124-13b-instruct",
        "score_percent": 13.641333333333334,
        "parameters_b_from_identifier": 13,
        "tasks": {
          "AMPS_Hard": 25,
          "math_comp": 2.083,
          "olympiad": 13.841
        },
        "shafu_difference_points": 66.92394454646453,
        "size_label": "13B",
        "parameter_status": "model_identifier",
        "size_note": "型号标注参数量"
      }
    ],
    "parameter_sources": [
      {
        "model_ids": [
          "deepseek-r1"
        ],
        "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1",
        "total_parameters_b": 671,
        "activated_parameters_b": 37,
        "weights": "open",
        "license": "MIT"
      },
      {
        "model_ids": [
          "deepseek-v3"
        ],
        "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3",
        "total_parameters_b": 671,
        "activated_parameters_b": 37,
        "weights": "open",
        "license": "DeepSeek Model License"
      },
      {
        "model_ids": [
          "deepseek-v3-0324"
        ],
        "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324",
        "total_parameters_b": 671,
        "activated_parameters_b": 37,
        "weights": "open",
        "license": "MIT"
      },
      {
        "model_ids": [
          "phi-4"
        ],
        "url": "https://huggingface.co/microsoft/phi-4",
        "total_parameters_b": 14,
        "source_note": "Use vendor-declared model scale; Hugging Face tensor-file counts may round to 15B."
      },
      {
        "model_ids": [
          "phi-3-small-8k-instruct"
        ],
        "url": "https://huggingface.co/microsoft/Phi-3-small-8k-instruct",
        "total_parameters_b": 7,
        "source_note": "Technical Specifications / Architecture"
      },
      {
        "model_ids": [
          "phi-3-mini-4k-instruct"
        ],
        "url": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct",
        "total_parameters_b": 3.8,
        "source_note": "Technical Specifications / Architecture"
      },
      {
        "model_ids": [
          "phi-3-mini-128k-instruct"
        ],
        "url": "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/blob/main/README.md",
        "total_parameters_b": 3.8,
        "source_note": "Model Summary"
      },
      {
        "model_ids": [
          "mistral-large-2411"
        ],
        "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411",
        "total_parameters_b": 123,
        "source_note": "Model Card introduction"
      },
      {
        "model_ids": [
          "mistral-small-2501"
        ],
        "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501",
        "total_parameters_b": 24,
        "source_note": "Model Card introduction"
      },
      {
        "model_ids": [
          "mistral-small-2503"
        ],
        "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503",
        "total_parameters_b": 24,
        "source_note": "Model Card introduction"
      },
      {
        "model_ids": [
          "command-r-08-2024"
        ],
        "url": "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024",
        "total_parameters_b": 32,
        "source_note": "Current official model card explicitly gives 32B. Older third-party copies still say 35B; do not copy 35B into this table."
      },
      {
        "model_ids": [
          "command-r-plus-08-2024"
        ],
        "url": "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024",
        "total_parameters_b": 104,
        "source_note": "Model Summary"
      }
    ],
    "verified_larger_count": 22,
    "weight_files": {
      "schema_version": 1,
      "checked_at": "2026-11-12T04:19:42.047Z",
      "models": {
        "qwq-32b": {
          "model_id": "qwq-32b",
          "repository": "Qwen/QwQ-32B",
          "file_bytes": 65527841688,
          "weight_gb": 65.527841688,
          "format": "safetensors",
          "precision": "BF16",
          "revision": "976055f8c83f394f35dbd3ab09a285a984907bd0",
          "file_count": 14,
          "source_url": "https://huggingface.co/Qwen/QwQ-32B/tree/976055f8c83f394f35dbd3ab09a285a984907bd0",
          "status": "verified",
          "note": "仅合计官方 model.safetensors.index.json 的 weight_map 引用的唯一 safetensors 权重文件（含文件头）；排除配置、分词器、索引、其他格式及重复权重。GB=10^9字节，不是运行内存/显存需求；历史评测服务实际精度及权重修订未锁定，此为官方发布文件体积参考。",
          "index_url": "https://huggingface.co/Qwen/QwQ-32B/resolve/976055f8c83f394f35dbd3ab09a285a984907bd0/model.safetensors.index.json",
          "metadata_url": "https://huggingface.co/api/models/Qwen/QwQ-32B/revision/976055f8c83f394f35dbd3ab09a285a984907bd0?blobs=true"
        },
        "deepseek-v3-0324": {
          "model_id": "deepseek-v3-0324",
          "repository": "deepseek-ai/DeepSeek-V3-0324",
          "file_bytes": 688586727753,
          "weight_gb": 688.586727753,
          "format": "safetensors",
          "precision": "FP8 mixed (F8_E4M3 + BF16 + F32)",
          "revision": "e9b33add76883f293d6bf61f6bd89b497e80e335",
          "file_count": 163,
          "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/tree/e9b33add76883f293d6bf61f6bd89b497e80e335",
          "status": "verified",
          "note": "仅合计官方 model.safetensors.index.json 的 weight_map 引用的163个唯一权重文件（含文件头及缩放张量）；模型API blobs大小与同修订tree API大小逐文件一致。官方为FP8混合精度；索引metadata.total_size=1369062772000与实际文件字节合计不一致，因此不以该字段充当下载体积。GB=10^9字节，不是运行内存/显存需求；历史评测服务实际精度及权重修订未锁定，此为官方发布文件体积参考。",
          "index_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/resolve/e9b33add76883f293d6bf61f6bd89b497e80e335/model.safetensors.index.json",
          "metadata_url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-0324/revision/e9b33add76883f293d6bf61f6bd89b497e80e335?blobs=true",
          "tree_metadata_url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V3-0324/tree/e9b33add76883f293d6bf61f6bd89b497e80e335?recursive=true&expand=false"
        },
        "deepseek-r1-distill-llama-70b": {
          "model_id": "deepseek-r1-distill-llama-70b",
          "repository": "deepseek-ai/DeepSeek-R1-Distill-Llama-70B",
          "file_bytes": 141107497606,
          "weight_gb": 141.107497606,
          "format": "safetensors",
          "precision": "BF16",
          "revision": "b1c0b44b4369b597ad119a196caf79a9c40e141e",
          "file_count": 17,
          "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/tree/b1c0b44b4369b597ad119a196caf79a9c40e141e",
          "status": "verified",
          "note": "仅合计官方 model.safetensors.index.json 的 weight_map 引用的唯一 safetensors 权重文件（含文件头）；排除配置、分词器、索引、其他格式及重复权重。GB=10^9字节，不是运行内存/显存需求；历史评测服务实际精度及权重修订未锁定，此为官方发布文件体积参考。",
          "index_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/resolve/b1c0b44b4369b597ad119a196caf79a9c40e141e/model.safetensors.index.json",
          "metadata_url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/revision/b1c0b44b4369b597ad119a196caf79a9c40e141e?blobs=true"
        },
        "phi-4": {
          "model_id": "phi-4",
          "repository": "microsoft/phi-4",
          "file_bytes": 29319042992,
          "weight_gb": 29.319042992,
          "format": "safetensors",
          "precision": "BF16",
          "revision": "2db69c1c3e91a05d2c64a3185acfbaf36f744e25",
          "file_count": 6,
          "source_url": "https://huggingface.co/microsoft/phi-4/tree/2db69c1c3e91a05d2c64a3185acfbaf36f744e25",
          "status": "verified",
          "note": "仅合计官方 model.safetensors.index.json 的 weight_map 引用的唯一 safetensors 权重文件（含文件头）；排除配置、分词器、索引、其他格式及重复权重。GB=10^9字节，不是运行内存/显存需求；历史评测服务实际精度及权重修订未锁定，此为官方发布文件体积参考。",
          "index_url": "https://huggingface.co/microsoft/phi-4/resolve/2db69c1c3e91a05d2c64a3185acfbaf36f744e25/model.safetensors.index.json",
          "metadata_url": "https://huggingface.co/api/models/microsoft/phi-4/revision/2db69c1c3e91a05d2c64a3185acfbaf36f744e25?blobs=true"
        },
        "mistral-large-2411": {
          "model_id": "mistral-large-2411",
          "repository": "mistralai/Mistral-Large-Instruct-2411",
          "file_bytes": 245220233776,
          "weight_gb": 245.220233776,
          "format": "safetensors",
          "precision": "BF16",
          "revision": "ba78820945ae22361b0274cf0ae6d696c967c1a4",
          "file_count": 51,
          "source_url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411/tree/ba78820945ae22361b0274cf0ae6d696c967c1a4",
          "status": "verified",
          "note": "仅合计官方 model.safetensors.index.json 的 weight_map 引用的唯一 safetensors 权重文件（含文件头）；排除配置、分词器、索引、其他格式及重复权重。GB=10^9字节，不是运行内存/显存需求；历史评测服务实际精度及权重修订未锁定，此为官方发布文件体积参考。 此仓库同时含 consolidated 另一套权重，本值仅计 Transformers model-* 这一套。",
          "index_url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411/resolve/ba78820945ae22361b0274cf0ae6d696c967c1a4/model.safetensors.index.json",
          "metadata_url": "https://huggingface.co/api/models/mistralai/Mistral-Large-Instruct-2411/revision/ba78820945ae22361b0274cf0ae6d696c967c1a4?blobs=true"
        }
      }
    }
  },
  "exam_cases": {
    "gaokao": {
      "label": "2024 新高考 II 卷",
      "score": 150,
      "max_score": 150,
      "kind": "多轮开发汇总案例",
      "composition": "前 18 题 133 分 + 第 19 题独立重测 17 分",
      "note": "不同运行合计，不是单次整卷首轮成绩；分步评分细则尚未逐项核对。"
    },
    "university": {
      "label": "大学精选数学题",
      "graded_questions": 32,
      "full_credit_questions": 19,
      "normalized_max_score_per_question": 100,
      "kind": "精选题内部人工评分；不是完整试卷",
      "subjects": {
        "数学分析（一）": 4,
        "线性代数": 2,
        "多元微积分": 5,
        "概率论": 1,
        "抽象代数": 3,
        "最优化": 4
      }
    },
    "elementary": {
      "status": "已收录原卷，评分待核验",
      "verified_full_paper_score": null
    },
    "middle_school": {
      "status": "已收录试卷与题目转录，评分待核验",
      "verified_full_paper_score": null
    },
    "gaokao_2025_I": {
      "generated_questions": 19,
      "status": "尚未评分",
      "verified_full_paper_score": null
    }
  },
  "product": {
    "name": "傻福 SHAFU",
    "positioning": "本地轻量管家",
    "qq": {
      "implementation": "QQ 官方机器人私聊与群 @ 回复流程已实现",
      "quick_create_scope": "创建人私聊",
      "group_reply": "需要 QQ 平台群聊权限",
      "requires_running_local_service": true,
      "real_account_end_to_end_verified": false
    }
  },
  "light_model": {
    "name": "Qwen3-1.7B",
    "quantization": "Q8_0",
    "dplus": true,
    "file_bytes": 1834426016,
    "file_gb": 1.834426016,
    "default_context": 8192,
    "platform": "Windows x64",
    "entry_configuration": {
      "ram_gb": 8,
      "gpu_vram_gb": 4,
      "gpu_example": "GTX 1650 4GB",
      "cpu_example": "四核 x64",
      "status": "estimated_unverified_not_minimum"
    },
    "note": "轻量档也已加入 D+ 技术；公开数学实验成绩仍来自量化 9B 的 D+ 技术配置。模型文件大小不等于内存或显存需求；2–3GB 显存不能保证可用，取决于上下文、可用显存、显卡兼容性和驱动。CPU 模式尚未完成验收。"
  }
}