Arena Agent · 任务确认完成
用户在会话里明确确认「任务完成了」的比例 完整榜单 53 个模型，样本量 1,461,754 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 Sonnet 5.5 (Max) | +20.5 |
2 | Claude Opus 5.5 (Max) | +18.5 |
3 | GPT 6.1 Sol (Max) | +15.9 |
4 | Claude Fable 5.1 (Max) | +15.4 |
5 | Gemini 4 Argon (High) | +14.7 |
6 | Claude Opus 5 (Max) | +12.7 |
7 | GPT 6 Astra (Max) | +12.5 |
8 | MiMo V2.6 Pro | +8.6 |
9 | Deepseek V4.1 Flash (Max) | +7.4 |
10 | GPT 6 Sol (Max) | +7.3 |
11 | MiMo V2.6 Flash | +6.7 |
12 | Kimi K3 (Max) | +6.1 |
13 | Hy4 preview | +6.0 |
14 | Grok 4.7 (xHigh) | +5.7 |
15 | Claude Fable 5 (High) | +5.6 |
16 | Gemini 3.8 Flash (High) | +5.2 |
17 | Qwen3.8 Max | +4.9 |
18 | Muse Spark 1.3 (Max) | +4.7 |
19 | GPT 5.6 Sol (xHigh) | +4.4 |
20 | Claude Opus 4.8 (High) | +4.2 |
21 | GLM 5.3 Flash | +4.0 |
22 | GLM 5.3 (Max) | +3.9 |
23 | GLM 5.2 (Max) | +3.8 |
24 | Qwen3.8 Flash Next | +2.6 |
25 | Mistral Large 4 | +2.1 |
26 | Claude Sonnet 5 (High) | +1.9 |
27 | Claude Haiku 5.5 (High) | +1.3 |
28 | GPT 6 Luna (Max) | +1.1 |
29 | Grok 4.5 | +0.7 |
30 | Qwen 3.8 27B | -0.3 |
31 | Step 5 Preview | -0.4 |
32 | Grok 4.6 (xHigh) | -0.6 |
33 | Gemini 3.7 Flash (High) | -0.8 |
34 | DeepSeek V4 Pro (High) (0813) | -1.4 |
35 | GPT 5.5 (xHigh) | -1.7 |
36 | GPT 5.6 Terra (xHigh) | -4.2 |
37 | GPT 5.4 (High) | -4.4 |
38 | Muse Spark 1.1 | -4.6 |
39 | Gemini 3.6 Flash (High) | -5.7 |
40 | GPT 5.6 Luna (xHigh) | -5.9 |
41 | GPT 5.5 | -6.0 |
42 | Muse Spark 1.2 (xHigh) | -7.7 |
43 | Qwen3.7 Plus | -7.8 |
44 | Qwen3.7 Max | -8.6 |
45 | Mimo V2.5 Pro | -8.8 |
46 | Gemini 3.1 Pro Preview | -10.0 |
47 | Hy3 | -10.4 |
48 | Minimax M3 | -13.6 |
49 | Mistral Medium 3.5 | -15.7 |
50 | Nemotron 3 Ultra | -17.0 |
51 | Inkling | -21.8 |
52 | Inkling Small | -24.5 |
53 | Solar Pro 4 | -26.0 |
数据更新: 2026-10-14数据来源: Arena AI ↗