Riemann-bench: leaderboard
The set holds 25 research-level maths problems contributed by working mathematicians from their own research, each with a single closed-form answer checked automatically with no partial credit. Surge AI keeps the problems private and runs models as agents with a code interpreter and web search, and the board reports the pass rate over repeated runs of each problem. Higher is better. Source: arxiv.org/abs/2604.06802.
Metric: Score (%). Source: surgehq.ai. Saturation forecast: Around July 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 74.4 | #19 (GPT-5.6 Sol) |
| 2 | GPT-6 Astra (Max) | 72 | #6 (GPT-6 Astra) |
| 3 | GPT-5.6 Terra (Max) | 66.4 | #41 (GPT-5.6 Terra) |
| 4 | Claude Fable 5.1 (Max) | 65.6 | #9 (Claude Fable 5.1) |
No result here: #7 GPT-6.1 Sol, #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=riemann-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.