{
  "schema_version": 1,
  "id": "evt-2024-deepseek-prover-v1-5",
  "slug": "deepseek-prover-v1-5-lean-proof-search",
  "title": {
    "en": "DeepSeek-Prover-V1.5 combines proof-assistant feedback, reinforcement learning and tree search",
    "zh-CN": "DeepSeek-Prover-V1.5 结合证明助手反馈、强化学习与树搜索"
  },
  "dates": {
    "event": "2024-08-15",
    "first_public_claim": "2024-08-15",
    "paper_release": "2024-08-15",
    "artifact_release": null,
    "independently_verified": null,
    "added_to_chronicle": "2026-09-10",
    "last_updated": "2026-09-10"
  },
  "event_types": [
    "system",
    "proof",
    "formalization"
  ],
  "significance": {
    "tier": "H2",
    "rationale": {
      "en": "DeepSeek-Prover-V1.5 made proof-assistant feedback a training signal through reinforcement learning and combined it with Monte Carlo tree search, while releasing the 7B model and proof-search implementation openly.",
      "zh-CN": "DeepSeek-Prover-V1.5 将证明助手反馈直接用作强化学习训练信号，并与蒙特卡洛树搜索结合，同时公开 7B 模型和证明搜索实现。"
    }
  },
  "mathematical_novelty": {
    "type": "benchmark_result"
  },
  "interfaces": [
    "formal",
    "symbolic"
  ],
  "claim": {
    "en": "DeepSeek-Prover-V1.5 reported 63.5% success on miniF2F-test and 25.3% on ProofNet using reinforcement learning from proof-assistant feedback plus RMaxTS proof search, improving on the earlier DeepSeek-Prover line.",
    "zh-CN": "DeepSeek-Prover-V1.5 通过证明助手反馈强化学习与 RMaxTS 证明搜索，在 miniF2F-test 上报告 63.5% 成功率，在 ProofNet 上报告 25.3%，较此前 DeepSeek-Prover 系列进一步提升。"
  },
  "summary": {
    "en": "The model was specialized for Lean 4 formal mathematics, refined with reinforcement learning from Lean feedback and paired with an intrinsic-reward Monte Carlo tree-search variant to explore diverse proof paths. DeepSeek released model weights and code for reproducing the proof-search setup.",
    "zh-CN": "该模型针对 Lean 4 形式数学进行专门训练，使用 Lean 反馈进行强化学习，并配合带内在奖励的蒙特卡洛树搜索变体探索多样证明路径。DeepSeek 公开了模型权重和复现实验的代码。"
  },
  "why_it_matters": {
    "en": "The system is a clear bridge from static proof generation toward learned agents that train and search against a formal verifier, while remaining small enough and open enough to become a widely reusable research baseline.",
    "zh-CN": "该系统清晰连接了静态证明生成与“面向形式验证器进行训练和搜索”的学习式 Agent，同时凭借较小规模与开放发布成为可广泛复用的研究基线。"
  },
  "ai_role": {
    "level": "ai_primary_human_verified",
    "description": {
      "en": "The language model generated Lean proofs and tree-search actions, while Lean supplied correctness feedback for both training and inference-time verification.",
      "zh-CN": "语言模型生成 Lean 证明与树搜索动作，Lean 同时为训练和推理阶段提供正确性反馈与验证。"
    }
  },
  "human_contribution": {
    "description": {
      "en": "Researchers assembled formal training data, designed reinforcement-learning and RMaxTS procedures, trained the models and selected benchmark protocols.",
      "zh-CN": "研究者构建形式训练数据，设计强化学习与 RMaxTS 流程，训练模型并选择评测协议。"
    }
  },
  "systems": [
    "deepseek-prover-v1-5",
    "lean"
  ],
  "organizations": [
    "deepseek-ai"
  ],
  "people": [
    "huajian-xin",
    "z-z-ren",
    "junxiao-song",
    "zhihong-shao",
    "fuli-luo"
  ],
  "problems": [
    "minif2f",
    "proofnet"
  ],
  "methods": [
    "reinforcement-learning-from-proof-feedback",
    "monte-carlo-tree-search",
    "formal-theorem-proving",
    "proof-search"
  ],
  "verification": {
    "evidence_level": "E3",
    "status": "paper_released",
    "formal_assurance": "machine_checked",
    "history": [
      {
        "date": "2024-08-15",
        "status": "paper_released",
        "evidence_level": "E2",
        "note": {
          "en": "The paper reported the training method and benchmark results.",
          "zh-CN": "论文公开了训练方法与基准结果。"
        }
      },
      {
        "date": "2024-08-15",
        "status": "paper_released",
        "evidence_level": "E3",
        "note": {
          "en": "Model weights, Lean integration and proof-search code were released publicly.",
          "zh-CN": "模型权重、Lean 集成与证明搜索代码被公开发布。"
        }
      }
    ]
  },
  "sources": [
    {
      "id": "arxiv-deepseek-prover-v1-5-2408.08152",
      "tier": "S1",
      "type": "paper",
      "title": "DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search",
      "url": "https://arxiv.org/abs/2408.08152",
      "primary": true,
      "published_date": "2024-08-15",
      "accessed_at": "2026-09-10"
    },
    {
      "id": "github-deepseek-prover-v1-5",
      "tier": "S1",
      "type": "repository",
      "title": "DeepSeek-Prover-V1.5 repository",
      "url": "https://github.com/deepseek-ai/DeepSeek-Prover-V1.5",
      "primary": true,
      "published_date": null,
      "accessed_at": "2026-09-10"
    }
  ],
  "artifacts": [
    {
      "id": "deepseek-prover-v1-5-repository",
      "kind": "code",
      "title": "DeepSeek-Prover-V1.5 models and prover code",
      "url": "https://github.com/deepseek-ai/DeepSeek-Prover-V1.5"
    }
  ],
  "relationships": {
    "predecessors": [
      "evt-2020-gpt-f",
      "evt-2021-minif2f",
      "evt-2023-leandojo"
    ],
    "successors": [
      "evt-2026-alphaproof-nexus"
    ],
    "related": [
      "evt-2023-copra",
      "evt-2024-alphaproof-alphageometry2-imo"
    ]
  },
  "timeline": {
    "featured": true,
    "significance": "primary"
  },
  "tags": [
    "lean",
    "theorem-proving",
    "reinforcement-learning",
    "tree-search",
    "open-model",
    "proof-feedback"
  ],
  "editorial": {
    "created_at": "2026-09-10",
    "last_reviewed": "2026-09-11",
    "reviewers": [
      "Charlie-Wang-03"
    ]
  },
  "corrections": []
}