LMArena Agent Arena Bash Reliability
Agents
Signed Agent Arena diagnostic for successful bash commands, excluding failures outside the model's control. Atlas retains the publisher's raw bash_reliability_no_prereq sign and scales by 100; this baseline-relative signal can be negative and is not an absolute command-success probability.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 15.7 |
| Sonnet 5 | 15.5 |
| Claude Sonnet 5.5 | 13.1 |
| GPT-5.5 | 13.0 |
| GPT-5.6 Sol | 12.0 |
| Claude Fable 5 | 12.0 |
| GPT-6 Sol | 11.4 |
| Claude Opus 5 | 11.2 |
| Gemini 3.8 Flash | 10.9 |
| GPT-6 Luna | 10.8 |
| Claude Fable 5.1 | 10.6 |
| GPT-5.4 | 10.1 |
| Grok 4.6 | 9.5 |
| Grok 4.7 | 9.4 |
| GPT-6 Astra | 9.2 |
| Muse Wonder Full | 7.7 |
| Claude Haiku 5.5 | 7.3 |
| GPT-5.6 Luna | 7.0 |
| Muse Wonder | 6.9 |
| Gemini 3.7 Flash | 6.7 |
| DeepSeek V4.1 Flash 1017 | 5.2 |
| Qwen3.7-Max (source unspecified) | 4.5 |
| MiMo-V2.6-Pro | 3.7 |
| GPT-6.1 Sol | 3.4 |
| Inkling-Small | 2.9 |
| Gemini 3.1 Pro Preview | 2.6 |
| Muse Spark 1.3 | 2.2 |
| DeepSeek V4 Pro 0813 | 2.2 |
| Qwen3.7-Plus (source unspecified) | 1.7 |
| GPT-5.6 Terra | 1.6 |
| Mistral Large 4 Preview | 1.1 |
| Inkling | 0.7 |
| Qwen3.8-Max | 0.1 |
| MiMo-V2.5-Pro | -0.1 |
| DeepSeek V4.1 Flash | -0.4 |
| MiniMax M3 | -0.4 |
| Muse Spark 1.1 | -1.8 |
| MiMo-V2.6-Flash | -2.1 |
| Hy4 Preview | -2.2 |
| Kimi K3 | -2.5 |