AlignBench: leaderboard
683 Chinese queries in eight categories with human-verified reference answers, from Tsinghua KEG and Zhipu AI (2023); an LLM judge with chain-of-thought rates each response against the reference.
Metric: Score (%). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 Preview (1106) | 80.1 | #418 |
| 2 | GPT-4 (0613) | 75.3 | #491 |
| 3 | GPT-3.5 Turbo (0613) | 60.8 | #1118 |
| 4 | Qwen-14B-Chat | 57.2 | #1169 |
| 5 | Baichuan2-13B-Chat | 52.5 | #1255 |
| 6 | chatglm3-6B | 49.7 | #1439 |
| 7 | Baichuan2-7B-Chat | 49.7 | #1420 |
| 8 | internlm-20B | 49.6 | #1296 |
| 9 | Qwen-7B Chat | 49.1 | #1380 |
| 10 | ChatGLM2 6B | 44.8 | #1498 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=alignbench · How It Works · Data refreshed daily, snapshot 2026-10-20.