MATH (CoT): leaderboard
12,500 competition maths problems (5,000 test) from AMC 10/12, AIME and more across seven subjects and five levels, UC Berkeley (NeurIPS 2021); chain-of-thought, exact match on the boxed answer.
Metric: Score (%). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Best Expert | 90 | |
| 2 | Llama 3.1 405B Instruct | 73.8 | #457 |
| 3 | Llama 3.1 70B Instruct | 68 | #561 |
| 4 | Llama 3.1 8B Instruct | 51.9 | #1035 |
| 5 | Llama 3 70B Instruct | 51 | #634 |
| 6 | Llama 3 8B Instruct | 29.1 | #1129 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=math-cot · How It Works · Data refreshed daily, snapshot 2026-10-20.