AdvancedMathBench - ProverBench Qualifying Exam: leaderboard
ProverBench of AdvancedMathBench (Shanghai AI Laboratory, 2026) asks models for natural-language proofs of advanced mathematics problems. This board is the percent of the 45 doctoral qualifying-exam problems whose proof all 8 parallel passes of an expert-trained automatic verifier accept, at the highest available reasoning effort; higher is better.
Metric: Proof acceptance rate (%) on the 45 doctoral qualifying-exam problems (QE split): temperature 1.0, 64k maximum output tokens, highest available reasoning effort; each proof is accepted only if all 8 parallel passes of the authors' expert-trained automatic verifier accept it (pessimistic verification); higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 48.9 | #49 (GPT-5.5) |
| 2 | GPT-5.5 (High) | 46.1 | #49 (GPT-5.5) |
| 3 | DeepSeek V4 Pro (Max) | 40 | #188 (DeepSeek V4 Pro) |
| 4 | Claude Opus 4.8 (Max) | 40 | #56 (Claude Opus 4.8) |
| 5 | Qwen 3.5 397B A17B | 33.5 | #246 |
| 6 | GLM-5.2 (Max) | 28.9 | #138 (GLM-5.2) |
| 7 | GPT-5.2 (xHigh) | 26.7 | #176 (GPT-5.2) |
| 8 | Gemini 3.1 Pro (Preview) (High) | 17.8 | #94 (Gemini 3.1 Pro (Preview)) |
| 9 | GPT-OSS-120B (High) | 2.2 | #700 (GPT-OSS-120B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=advancedmathbench-proverbench-qualifying-exam · How It Works · Data refreshed daily, snapshot 2026-11-13.