Arena Agent · 可操控性
用户纠正、反驳之后,模型能改到点子上的比例 完整榜单 53 个模型,样本量 6,329,448 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 | Claude Opus 5.5 (Max) | +17.5 |
2 | Gemini 4 Argon (High) | +16.5 |
3 | Claude Fable 5.1 (Max) | +15.0 |
4 | GPT 6.1 Sol (Max) | +14.4 |
5 | GPT 6 Astra (Max) | +13.7 |
6 | GPT 6 Sol (Max) | +12.1 |
7 | Claude Sonnet 5.5 (Max) | +12.1 |
8 | Claude Fable 5 (High) | +10.3 |
9 | GPT 5.6 Sol (xHigh) | +8.5 |
10 | Claude Opus 4.8 (High) | +7.6 |
11 | Claude Opus 5 (Max) | +7.4 |
12 | GPT 5.4 (High) | +7.3 |
13 | Claude Sonnet 5 (High) | +6.6 |
14 | Claude Haiku 5.5 (High) | +6.1 |
15 | GPT 6 Luna (Max) | +6.1 |
16 | Muse Spark 1.3 (Max) | +5.4 |
17 | MiMo V2.6 Flash | +4.3 |
18 | GPT 5.5 (xHigh) | +3.4 |
19 | Grok 4.5 | +2.9 |
20 | GPT 5.6 Terra (xHigh) | +2.5 |
21 | Mistral Large 4 | +2.5 |
22 | Hy4 preview | +2.4 |
23 | Deepseek V4.1 Flash (Max) | +1.9 |
24 | Gemini 3.8 Flash (High) | +1.9 |
25 | MiMo V2.6 Pro | +1.9 |
26 | Qwen3.8 Max | +1.9 |
27 | GLM 5.2 (Max) | +1.3 |
28 | Grok 4.6 (xHigh) | +1.2 |
29 | Grok 4.7 (xHigh) | +0.9 |
30 | Step 5 Preview | +0.8 |
31 | Hy3 | +0.8 |
32 | DeepSeek V4 Pro (High) (0813) | +0.6 |
33 | GLM 5.3 (Max) | +0.5 |
34 | Kimi K3 (Max) | +0.3 |
35 | Qwen3.8 Flash Next | -0.9 |
36 | GPT 5.5 | -1.0 |
37 | GPT 5.6 Luna (xHigh) | -1.3 |
38 | GLM 5.3 Flash | -2.9 |
39 | Qwen 3.8 27B | -3.3 |
40 | Gemini 3.7 Flash (High) | -3.4 |
41 | Gemini 3.1 Pro Preview | -5.4 |
42 | Gemini 3.6 Flash (High) | -7.4 |
43 | Muse Spark 1.2 (xHigh) | -8.5 |
44 | Minimax M3 | -8.5 |
45 | Qwen3.7 Plus | -9.0 |
46 | Nemotron 3 Ultra | -9.9 |
47 | Qwen3.7 Max | -10.2 |
48 | Mistral Medium 3.5 | -10.9 |
49 | Muse Spark 1.1 | -11.0 |
50 | Mimo V2.5 Pro | -11.5 |
51 | Solar Pro 4 | -14.1 |
52 | Inkling | -15.8 |
53 | Inkling Small | -19.1 |
数据更新: 2026-10-14数据来源: Arena AI ↗