AI搜索能力排行榜
基于Arena Search Elo评分，找到搜索和信息检索最强的大模型。
分数怎么算：分数 = (Elo − 900) ÷ (全站最高 Elo − 900) × 100，与其他 Arena 榜共用一把尺子。
Arena AI 数据更新于 2026-10-14，每天北京时间凌晨自动刷新
| Rank | Model | Score |
|---|---|---|
1 | Qwen3.8 Max (0902) | 39.6 |
2 | GPT-5.6 Sol (xhigh) | 39.6 |
3 | Claude Opus 4.6 (non-reasoning, high) | 38.6 |
4 | Claude Fable 5.1 (high with fallback) | 38.0 |
5 | GPT-5.5 (high) | 38.0 |
6 | Claude Opus 5 (high) | 37.5 |
7 | GPT-6 Astra (high) | 36.6 |
8 | GPT-6.1 Sol (high) | 36.6 |
9 | Claude Opus 4.7 (max) | 36.2 |
10 | Claude Fable 5.1 (max with fallback) | 36.1 |
11 | Claude Sonnet 4.6 (non-reasoning, high) | 35.2 |
12 | Claude Opus 5.5 (medium with fallback) | 34.3 |
13 | Gemini 3.1 Pro Preview | 33.9 |
14 | Gemini 3 Pro Preview (low) | 33.6 |
15 | Claude Opus 4.8 (max) | 33.4 |
16 | GPT-5.4 (low) | 32.8 |
17 | Gemini 3 Flash (non-reasoning) | 32.5 |
18 | Claude Sonnet 5 (high) | 31.9 |
19 | Claude Opus 4.5 | 30.7 |
20 | GPT-6 Sol (high) | 30.2 |
21 | GPT-6 Luna (high) | 28.5 |
22 | o3 | 26.9 |
23 | Gemini 2.5 Pro | 26.6 |
24 | GPT-5 (low) | 25.6 |
25 | GPT-4o (Nov) | 11.7 |