MathlibLemma: leaderboard
Metric: Success@2 (%): share of the 4,028 non-trivial MathlibLemma Lean 4 lemma statements (folklore lemmas generated from Mathlib seed files, type-checked, aesop-trivial ones removed) that the model proves closed-book, accepted only if Lean checks the proof and it passes the proof-bypass screen, within one attempt plus up to two compiler-feedback repair rounds; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 (Low) | 19.39 | #131 (GPT-5.1) |
| 2 | GPT-5.1 (Non-reasoning) | 15.69 | #131 (GPT-5.1) |
| 3 | DeepSeek R1 Distill Qwen 32B | 6.98 | #640 |
| 4 | DeepSeek R1 Distill Llama 70B | 6.68 | #582 |
| 5 | Qwen 3 235B A22B 2507 (Thinking) | 2.46 | #253 (Qwen 3 235B A22B 2507) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mathliblemma · How It Works · Data refreshed daily, snapshot 2026-10-11.