AlpacaEval 2.0: leaderboard
Automated instruction-following evaluation using length-controlled win rates against GPT-4 Turbo. 805 prompts scored by an LLM judge. Fast, cheap, and highly correlated with Chatbot Arena.
Metric: LC Win Rate (%). Source: tatsu-lab.github.io. Saturation forecast: Estimated already saturated. 223 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemma-2-9B-it-SimPO | 72.35 | #580 |
| 2 | gemma-2-9B-it-DPO | 67.66 | #678 |
| 3 | GPT-4o (2024-05-13) | 57.46 | #325 |
| 4 | GPT-4 Turbo | 55.02 | #404 |
| 5 | Claude 3.5 Sonnet (20240620) | 52.37 | #299 |
| 6 | GPT-4o Mini (2024-07-18) | 50.73 | #558 |
| 7 | Llama-3-Instruct-8B-SimPO | 44.65 | #1048 |
| 8 | Qwen 1.5 110B Chat | 43.91 | #691 |
| 9 | Nanbeige2-16B-Chat | 40.59 | #978 |
| 10 | Claude 3 Opus (20240229) | 40.51 | #412 |
| 11 | Llama 3.1 405B Instruct | 39.26 | #457 |
| 12 | GPT-4 | 38.13 | #438 |
| 13 | Qwen 2 72B Instruct | 38.07 | #576 |
| 14 | Llama 3.1 70B Instruct | 38.06 | #561 |
| 15 | Qwen 1.5 72B Chat | 36.57 | #776 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=alpacaeval-2-0 · How It Works · Data refreshed daily, snapshot 2026-10-20.