{
  "schema_version": 1,
  "id": "evt-2024-deepseekmath",
  "slug": "deepseekmath-grpo-open-math-model",
  "title": {
    "en": "DeepSeekMath couples large-scale math pretraining with GRPO",
    "zh-CN": "DeepSeekMath 将大规模数学预训练与 GRPO 结合"
  },
  "dates": {
    "event": "2024-02-05",
    "first_public_claim": "2024-02-05",
    "paper_release": "2024-02-05",
    "artifact_release": null,
    "independently_verified": null,
    "added_to_chronicle": "2026-09-10",
    "last_updated": "2026-09-11"
  },
  "event_types": [
    "system"
  ],
  "significance": {
    "tier": "H2",
    "rationale": {
      "en": "DeepSeekMath paired a strong open mathematical corpus pipeline with GRPO, a reinforcement-learning method that later became influential in reasoning-model training.",
      "zh-CN": "DeepSeekMath 将高质量开放数学语料管线与 GRPO 强化学习结合，后者随后成为推理模型训练中的重要方法。"
    }
  },
  "mathematical_novelty": {
    "type": "benchmark_result"
  },
  "interfaces": [
    "informal",
    "symbolic"
  ],
  "claim": {
    "en": "DeepSeekMath-7B reached 51.7% on MATH without external tools or voting and introduced Group Relative Policy Optimization (GRPO).",
    "zh-CN": "DeepSeekMath-7B 在不使用外部工具或投票的情况下于 MATH 达到 51.7%，并提出 Group Relative Policy Optimization（GRPO）。"
  },
  "summary": {
    "en": "The model continued pretraining on 120B math-related tokens and then used supervised and reinforcement-learning stages for mathematical reasoning.",
    "zh-CN": "该模型在 120B 数学相关 token 上继续预训练，再通过监督与强化学习阶段提升数学推理。"
  },
  "why_it_matters": {
    "en": "It linked open math-model scaling with a verifier-friendly RL recipe that influenced subsequent reasoning systems.",
    "zh-CN": "它把开放数学模型扩展与适合可验证任务的强化学习方法连接起来，并影响了后续推理系统。"
  },
  "ai_role": {
    "level": "computation_only",
    "description": {
      "en": "The trained model performs competition-level mathematical reasoning.",
      "zh-CN": "训练后的模型执行竞赛级数学推理。"
    }
  },
  "human_contribution": {
    "description": {
      "en": "Researchers built the data pipeline, model training recipe and GRPO optimization method.",
      "zh-CN": "研究者构建数据管线、模型训练方案与 GRPO 优化方法。"
    }
  },
  "systems": [
    "deepseekmath"
  ],
  "organizations": [
    "deepseek"
  ],
  "people": [],
  "problems": [
    "math-benchmark"
  ],
  "methods": [
    "grpo",
    "continued-pretraining",
    "reinforcement-learning"
  ],
  "verification": {
    "evidence_level": "E3",
    "status": "paper_released",
    "formal_assurance": "artifact_available",
    "history": [
      {
        "date": "2024-02-05",
        "status": "paper_released",
        "evidence_level": "E2",
        "note": {
          "en": "The Chronicle records this event from the cited primary or official source; the status does not imply independent Chronicle verification.",
          "zh-CN": "Chronicle 依据所引 primary / official source 记录该事件；该状态不表示 Chronicle 已完成独立验证。"
        }
      },
      {
        "date": "2026-09-11",
        "status": "paper_released",
        "evidence_level": "E3",
        "note": {
          "en": "This corpus refresh added a directly inspectable public artifact maintained by the authors or releasing organization. E3 records artifact-level inspectability and does not imply independent verification of the underlying scientific or mathematical claim.",
          "zh-CN": "本轮 corpus 刷新增补了由作者或发布机构维护、可直接检查的公开产物。E3 表示证据已达到可检查 artifact 层级，并不等同于底层科学或数学主张已获得独立验证。"
        }
      }
    ]
  },
  "sources": [
    {
      "id": "arxiv-deepseekmath-2402.03300",
      "tier": "S1",
      "type": "paper",
      "title": "DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models",
      "url": "https://arxiv.org/abs/2402.03300",
      "primary": true,
      "published_date": "2024-02-05",
      "accessed_at": "2026-09-10"
    },
    {
      "id": "github-deepseek-math",
      "tier": "S1",
      "type": "repository",
      "title": "DeepSeekMath code, evaluation and model release repository",
      "url": "https://github.com/deepseek-ai/DeepSeek-Math",
      "primary": true,
      "published_date": null,
      "accessed_at": "2026-09-11"
    }
  ],
  "artifacts": [
    {
      "id": "deepseekmath-repository",
      "kind": "code",
      "title": "DeepSeekMath evaluation code and model links",
      "url": "https://github.com/deepseek-ai/DeepSeek-Math"
    }
  ],
  "relationships": {
    "predecessors": [
      "evt-2023-llemma",
      "evt-2023-math-shepherd"
    ],
    "successors": [],
    "related": []
  },
  "timeline": {
    "featured": true,
    "significance": "primary"
  },
  "tags": [
    "deepseek",
    "grpo",
    "open-model",
    "math-reasoning"
  ],
  "editorial": {
    "created_at": "2026-09-10",
    "last_reviewed": "2026-09-11",
    "reviewers": [
      "Charlie-Wang-03"
    ]
  },
  "corrections": []
}