TheoremQA: leaderboard
800 questions covering 350 theorems in math, physics, EE/CS and finance, with integer, float, list, boolean or option answers, from the University of Waterloo (2023); GPT-4 scored 51% at release.
Metric: Score (%). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 Turbo | 48.4 | #404 |
| 2 | Llama 3 70B Instruct | 42 | #634 |
| 3 | Mixtral 8x7B Instruct | 25.3 | #1170 |
| 4 | deepseek-llm-67B-base | 25.3 | #973 |
| 5 | Yi 34B (Base) | 23.2 | #972 |
| 6 | Llama 3 8B Instruct | 22.8 | #1129 |
| 7 | internlm2-20B | 19.5 | #943 |
| 8 | deepseek-llm-7B-base | 15.7 | #1443 |
| 9 | Yi 6B (Base) | 12 | #1472 |
| 10 | Llama 2 13B Base | 10.9 | #1445 |
| 11 | internlm2-7B | 7.8 | #1192 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=theoremqa · How It Works · Data refreshed daily, snapshot 2026-10-20.