Research map / Large language models / Agents, code and math
Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction
Yong Lin, Shange Tang, Bohan Lyu, Ziran Yang, Jui-Hui Chung, Haoyu Zhao, Lai Jiang, Yihan Geng, Jiawei Ge, Jingruo Sun, Jiayun Wu, Jiri Gesi and 8 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add verifier-guided self-correction (2 rounds) (ablation) | Goedel-Prover-V2-32B · MiniF2F test pass@32 | pass@32 (%): 88.1 -> 90.4 | Improved |
| Add verifier-guided self-correction (2 rounds) (ablation) | Goedel-Prover-V2-8B · MiniF2F test pass@32 | pass@32 (%): 84.6 -> 86.7 | Improved |
| Add verifier-guided self-correction (ablation) | Goedel-Prover-V2-32B · PutnamBench pass@32 | problems solved: 43 -> 57 | Improved |
| Remove compiler error messages from the self-correction prompt (ablation) | Goedel-Prover-V2 self-correction · MiniF2F pass@32 | pass@32 | Hurt |
| Keep previous chain-of-thought in the correction context (ablation) | Goedel-Prover-V2 self-correction · MiniF2F pass@32 | pass@32 | Improved |
| Extend context to 128k and allow 5 revision iterations (ablation) | Goedel-Prover-V2-32B self-correction · MiniF2F pass@32 | pass@32 (%): 90.4 -> 92.7 | Improved |
| Train formalizer with reasoning traces and expert iteration (Goedel-Formalizer-V2 vs Kimina-autoformalizer) | Formalizer · 300 Omni-math problems | formalization passes: 161 -> 228 | Improved |
| More RL steps (ablation) | Goedel-Prover-V2 · MiniF2F | pass@1 | Improved |
| More RL steps (ablation) | Goedel-Prover-V2 vanilla mode · MiniF2F | pass@N | No clear effect |
| More RL steps (ablation) | Goedel-Prover-V2 correction mode · MiniF2F | pass@N | Improved |
| Model averaging with the base model (increasing weight on base model) (ablation) | RL checkpoints · MiniF2F | pass@1 and pass@N | Mixed |
| Scale from 8B to 32B | Goedel-Prover-V2 · MiniF2F pass@32 | pass@32 (%): 84.6 -> 88.1 | Improved |
Closest papers by meaning
- ImProver: Agent-Based Automated Proof Optimization (ICLR 2025 · similarity 0.90)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.90)
- MPS-Prover: Advancing Stepwise Theorem Proving by Multi-Perspective Search and Data Curation (NeurIPS 2025 · similarity 0.90)
- ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis (ICML 2025 · similarity 0.89)
- Lean-STaR: Learning to Interleave Thinking and Proving (ICLR 2025 · similarity 0.89)
- LEGO-Prover: Neural Theorem Proving with Growing Libraries (ICLR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
