HELM Capabilities - MMLU-Pro: leaderboard
HELM Capabilities run of MMLU-Pro, a harder multi-domain knowledge benchmark scored by chain-of-thought accuracy.
Metric: COT correct. Source: crfm.stanford.edu. Saturation forecast: Around January 2027. 68 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 90.3 | #66 |
| 2 | Claude Opus 4 (20250514) (Thinking) | 87.5 | #154 (Claude Opus 4 (20250514)) |
| 3 | Claude Sonnet 4.5 | 86.9 | #142 |
| 4 | GPT-5 | 86.3 | #98 |
| 5 | Gemini 2.5 Pro (Preview 03-25) | 86.3 | #123 |
| 6 | O3 (2025-04-16) | 85.9 | #121 |
| 7 | Claude Opus 4 (20250514) | 85.9 | #154 |
| 8 | Grok 4 | 85.1 | #181 |
| 9 | Claude Sonnet 4 (20250514) | 84.3 | #213 |
| 10 | Claude Sonnet 4 (20250514) (Thinking) | 84.3 | #213 (Claude Sonnet 4 (20250514)) |
| 11 | GPT-5 Mini (2025-08-07) | 83.5 | #172 |
| 12 | O4 Mini (2025-04-16) | 82 | #179 |
| 13 | Kimi K2 | 81.9 | #243 |
| 14 | GPT-4.1 (2025-04-14) | 81.1 | #250 |
| 15 | Llama 4 Maverick Instruct FP8 | 81 | #414 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-capabilities-mmlu-pro · How It Works · Data refreshed daily, snapshot 2026-10-20.