AlpacaEval 2.0 — Standard win-rate %; LC-winrate sibling uses a distinct length-controlled metric, kept separate
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Granite 3.3 8B Instruct | IBM | 62.7 | 2 | 2026-08-23 |
| 2 | Granite 3.2 8B Instruct | IBM | 61.2 | 1 | 2026-06-11 |
| 3 | Granite 4.1 30B | IBM | 56.2 | 1 | 2026-06-15 |
| 4 | Hunyuan Large Inst. | Tencent | 51.8 | 1 | 2026-05-31 |
| 5 | DeepSeek-V2.5 | DeepSeek | 50.5 | 2 | 2026-08-23 |
| 6 | Granite 4.1 8B | IBM | 50.1 | 1 | 2026-06-15 |
| 7 | Granite 3.3 2B Instruct | IBM | 43.5 | 1 | 2026-06-11 |
| 8 | Llama 3.1 Instruct 405B | Meta | 39.3 | 1 | 2026-05-31 |
| 9 | Granite 3.2 2B Instruct | IBM | 34.5 | 1 | 2026-06-11 |
| 10 | Llama 3.1 70B Instruct | Meta | 34.3 | 1 | 2026-05-31 |
| 11 | Granite 3.1 8B Instruct | IBM | 30.3 | 1 | 2026-06-11 |
| 12 | Llama 3.1 8B Instruct | Meta | 27.2 | 1 | 2026-06-11 |
| 13 | Granite 3.1 2B Instruct | IBM | 27.2 | 1 | 2026-06-11 |
| 14 | DeepSeek-R1-Distill-Llama-8B | DeepSeek | 21.9 | 1 | 2026-06-11 |
| 15 | DeepSeek-R1-Distill-Qwen-7B | DeepSeek | 15.3 | 1 | 2026-06-11 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.