MA-ProofBench - Qualifying Exam: leaderboard
MA-ProofBench (2026) Pass@8 on the 100 Level II theorems, formalized from PhD qualifying exam problems in measure theory, complex, functional and real analysis: the share of theorems for which one of eight sampled Lean 4 proofs compiles and verifies, in percent; higher is better.
Metric: Pass@8 (%) on the 100 Level II (PhD qualifying exam) theorems, formal Lean 4 theorem proving: the model writes a complete Lean 4 proof of a formalized theorem, accepted only when the Lean compiler verifies it; Pass@8 over sampled proofs (8 for the proprietary models, 32 for the open models, estimated at 8); higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 5 | #94 (Gemini 3.1 Pro (Preview)) |
| 2 | GPT-5.5 (xHigh) | 5 | #49 (GPT-5.5) |
| 3 | Claude Sonnet 4.6 (High) | 3 | #161 (Claude Sonnet 4.6) |
| 4 | DeepSeek V3.2 (Thinking) | 1.85 | #406 (DeepSeek V3.2) |
| 5 | Qwen 3.5 397B A17B | 0 | #246 |
| 6 | GLM-5.1 | 0 | #179 |
| 7 | Qwen 3 235B A22B 2507 (Thinking) | 0 | #463 |
| 8 | GPT-OSS-120B (High) | 0 | #700 (GPT-OSS-120B) |
| 9 | nemotron-3-nano-30B-a3B | 0 | #1875 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ma-proofbench-qualifying-exam · How It Works · Data refreshed daily, snapshot 2026-11-13.