MixEval — leaderboard
Derives queries from real-world user interactions and matches them with similar existing benchmark queries. Achieves 0.96 correlation with Chatbot Arena while being much cheaper to run.
Metric: Score. Source: mixeval.github.io. Status: saturation imminent. 52 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 Preview | 72 |
| 2 | Llama 3.1 405B Instruct | 66.2 |
| 3 | GPT-4o (2024-05-13) | 64.7 |
| 4 | Claude 3 Opus | 63.5 |
| 5 | GPT-4 Turbo | 62.6 |
| 6 | Mistral Large 2 (Jul) | 57.4 |
| 7 | Yi Large (Preview) | 56.8 |
| 8 | Llama 3 70B Instruct | 55.9 |
| 9 | Claude 3 Sonnet | 54 |
| 10 | MAmmoTH2-8x7B-Plus | 51.8 |
| 11 | DeepSeek V2 | 51.7 |
| 12 | GPT-4o Mini | 51.6 |
| 13 | Command-R+ | 51.4 |
| 14 | Yi 1.5 34B Chat | 51.2 |
| 15 | Mistral Large | 50.3 |
Interactive version: theaggregate.ai/benchmark?slug=mixeval · How the rankings work · Data refreshed daily, snapshot 2026-07-22.