Arena Agent · 好评 vs 差评
用户明确表扬多于明确抱怨的程度 完整榜单 53 个模型,样本量 620,726 次观测。
综合总榜535 个行为信号加权后的总排名任务确认完成53用户在会话里明确确认「任务完成了」的比例好评 vs 差评53用户明确表扬多于明确抱怨的程度可操控性53用户纠正、反驳之后,模型能改到点子上的比例Bash Reliability53How often the model’s bash commands succeed. Errors outside its control aren’t counted. E.g. network issues, permission denied, etc.工具幻觉53调用根本不存在的工具的情况,分越高越少乱编
| Rank | Model | 信号分 |
|---|---|---|
1 | GPT 6.1 Sol (Max) | +35.5 |
2 | Claude Opus 5.5 (Max) | +32.4 |
3 | Gemini 4 Argon (High) | +30.9 |
4 | GPT 6 Astra (Max) | +29.3 |
5 | Claude Fable 5.1 (Max) | +28.0 |
6 | GPT 6 Sol (Max) | +22.0 |
7 | Claude Sonnet 5.5 (Max) | +18.9 |
8 | GPT 5.6 Sol (xHigh) | +15.5 |
9 | Claude Fable 5 (High) | +15.2 |
10 | Claude Opus 5 (Max) | +14.9 |
11 | Claude Opus 4.8 (High) | +14.0 |
12 | GLM 5.2 (Max) | +7.4 |
13 | GPT 5.5 (xHigh) | +6.7 |
14 | GLM 5.3 (Max) | +6.3 |
15 | Hy4 preview | +5.3 |
16 | Claude Sonnet 5 (High) | +5.2 |
17 | Gemini 3.8 Flash (High) | +4.8 |
18 | MiMo V2.6 Pro | +4.4 |
19 | Muse Spark 1.3 (Max) | +3.8 |
20 | Deepseek V4.1 Flash (Max) | +3.6 |
21 | Grok 4.7 (xHigh) | +3.4 |
22 | GPT 5.5 | +3.2 |
23 | MiMo V2.6 Flash | +2.7 |
24 | GPT 5.6 Terra (xHigh) | +2.6 |
25 | Kimi K3 (Max) | +2.4 |
26 | Qwen3.8 Max | +2.3 |
27 | GPT 6 Luna (Max) | +1.3 |
28 | DeepSeek V4 Pro (High) (0813) | +0.8 |
29 | Grok 4.6 (xHigh) | -0.5 |
30 | GPT 5.4 (High) | -1.1 |
31 | GLM 5.3 Flash | -1.1 |
32 | Gemini 3.1 Pro Preview | -1.3 |
33 | Mistral Large 4 | -1.9 |
34 | Grok 4.5 | -2.3 |
35 | Gemini 3.7 Flash (High) | -2.4 |
36 | Nemotron 3 Ultra | -2.5 |
37 | Qwen3.8 Flash Next | -3.2 |
38 | Step 5 Preview | -3.2 |
39 | Claude Haiku 5.5 (High) | -3.9 |
40 | GPT 5.6 Luna (xHigh) | -4.9 |
41 | Qwen 3.8 27B | -5.0 |
42 | Hy3 | -6.0 |
43 | Gemini 3.6 Flash (High) | -7.9 |
44 | Qwen3.7 Plus | -10.5 |
45 | Qwen3.7 Max | -10.5 |
46 | Muse Spark 1.1 | -11.0 |
47 | Muse Spark 1.2 (xHigh) | -11.4 |
48 | Mimo V2.5 Pro | -13.4 |
49 | Minimax M3 | -13.5 |
50 | Mistral Medium 3.5 | -15.6 |
51 | Solar Pro 4 | -20.9 |
52 | Inkling Small | -22.6 |
53 | Inkling | -22.9 |
数据更新: 2026-10-14数据来源: Arena AI ↗