LemmaBench (April 2026, DeepSeek-R1 Judge): leaderboard
Metric: Proof acceptance rate (%): share of the April 2026 batch's research-level lemmas (120 self-contained lemmas extracted by GPT-5.5 from 31 late-April arXiv preprints; DeepSeek-R1's rate is over 110) whose single generated natural-language proof the DeepSeek-R1 judge accepts as a whole-proof verdict; higher is better. Source: arxiv.org. Saturation forecast: Around June 2028. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 51.7 | #26 |
| 2 | Gemini 3.1 Pro (Preview) | 44.2 | #54 |
| 3 | DeepSeek R1 | 40.9 | #245 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=lemmabench-april-2026-deepseek-r1-judge · How It Works · Data refreshed daily, snapshot 2026-10-11.