AlpacaEval: leaderboard
Automatic instruction-following evaluator comparing model responses against a reference using GPT-4 judgments and length-controlled win rates.
Metric: Length-Controlled Win Rate (%). Source: tatsu-lab.github.io. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 58 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Mistral Medium | 91.54 | #792 |
| 2 | GPT-4 | 86.51 | #438 |
| 3 | GPT-4 (0314) | 85.33 | #574 |
| 4 | LMCocktail-10.7B-v1 | 84.78 | #1008 |
| 5 | tulu-2-dpo-70B | 84.26 | #1052 |
| 6 | Mistral 7B Instruct (v0.2) | 82.98 | #1333 |
| 7 | Mixtral 8x7B Instruct (v0.1) | 82.6 | #1073 |
| 8 | GPT-4 (0613) | 81.38 | #491 |
| 9 | tulu-2-dpo-13B | 81.24 | #1340 |
| 10 | tulu-2-dpo-7B | 77.85 | #1400 |
| 11 | Yi Chat 34B | 76.36 | #976 |
| 12 | zephyr-7B-beta | 76.29 | #1330 |
| 13 | GPT-3.5 Turbo (1106) | 75.56 | #961 |
| 14 | Claude 2 | 74.34 | #761 |
| 15 | Llama 2 70B Chat (HF) | 74.11 | #1264 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=alpacaeval · How It Works · Data refreshed daily, snapshot 2026-10-20.