Research map / Large language models / Agents, code and math
miniF2F: a cross-system benchmark for formal Olympiad-level mathematics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply GPT-f/PACT in Lean rather than GPT-f in Metamath | GPT-f theorem proving · miniF2F-test | Pass@1: 1.3% -> 24.6% | Helped |
| Apply GPT-f/PACT in Lean rather than GPT-f in Metamath | GPT-f theorem proving · miniF2F-test | Pass@8: 1.6% -> 29.2% | Helped |
| Use Lean GPT-f/PACT instead of the Lean tidy baseline | Lean theorem proving · miniF2F-test | Pass@1: 18.0% -> 24.6% | Helped |
| Increase tidy search iteration budget from i_max = 1 to i_max = 8 (ablation) | Lean tidy · miniF2F-test | solved statements: 23~/ 244 -> 44~/ 244 | Helped |
| Increase tidy search iteration budget beyond i_max = 8 (ablation) | Lean tidy · miniF2F-test | solved statements: 44~/ 244; 44~/ 244; 43~/ 244; 44~/ 244 | No clear effect |
| Use the search-based tidy baseline rather than Lean's built-in tidy tactic | Lean tidy · miniF2F-test | solved statements: 13~/ 244 -> 44~/ 244 | Helped |
| Expose high-level tactics in Lean rather than use Metamath's low-level proof steps | GPT-f theorem proving · miniF2F-test | Pass@1: 1.3% -> 24.6% | Helped |
Closest papers by meaning
- PutnamBench: Evaluating Neural Theorem-Provers on the Putnam Mathematical Competition (NeurIPS 2024 · similarity 0.90)
- miniF2F-Lean Revisited: Reviewing Limitations and Charting a Path Forward (NeurIPS 2025 · similarity 0.90)
- Proving Olympiad Inequalities by Synergizing LLMs and Symbolic Reasoning (ICLR 2025 · similarity 0.89)
- LEGO-Prover: Neural Theorem Proving with Growing Libraries (ICLR 2024 · similarity 0.89)
- miniCTX: Neural Theorem Proving with (Long-)Contexts (ICLR 2025 · similarity 0.89)
- ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis (ICML 2025 · similarity 0.86)
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.