AlignBench — Chinese LLM alignment benchmark; v1.1+GPT4.1-judge sibling kept separate (version+judge differ)
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen2.5 Instruct 72B | Alibaba | 81.6 | 1 | 2026-08-23 |
| 2 | DeepSeek-V2.5 | DeepSeek | 80.4 | 2 | 2026-08-23 |
| 3 | Qwen2 7B Instruct | Alibaba | 72.1 | 1 | 2026-08-23 |
| 4 | Hunyuan Large Inst. | Tencent | 8.3 | 1 | 2026-05-31 |
| 5 | Llama 3.1 Instruct 405B | Meta | 6 | 1 | 2026-05-31 |
| 6 | Llama 3.1 70B Instruct | Meta | 5.9 | 1 | 2026-05-31 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.