Epoch AI - Proofbench: leaderboard
Epoch AI ProofBench evaluates formal theorem proving: models must write Lean 4 proofs for vetted formalizations of graduate-level math theorems, scored by verifier-checked proof success rate.
Metric: Score. Source: epoch.ai. 47 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | claude-fable-5-1_max | 100 | |
| 2 | claude-opus-5-5_max | 100 | |
| 3 | claude-sonnet-5-5_max | 100 | |
| 4 | claude-opus-5_max | 99 | |
| 5 | gpt-6-astra_unknown | 99 | |
| 6 | gpt-6.1-sol_unknown | 99 | |
| 7 | gemini-4-argon_unknown | 99 | |
| 8 | claude-fable-5_max | 95 | |
| 9 | claude-haiku-5-5_max | 89 | |
| 10 | kimi-k3_unknown | 87 |
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-proofbench · How It Works · Data refreshed daily, snapshot 2026-10-20.