LMArena Agent Arena Work Steering Burden
Agents · 2026-08-14
Signed percentage-point Agent Arena Work category diagnostic signal for the steering_burden signal (displayed by the source as Steerability), measuring response to user corrections, recomputed on sessions tagged as office work such as document creation, professional research, planning and professional writing. Values may be negative relative to the category baseline; reported confidence intervals are not standard errors.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 18.4 |
| Claude Fable 5.1 | 17.4 |
| Gemini 4 Argon | 14.9 |
| GPT-6.1 Sol | 14.7 |
| GPT-6 Astra | 14.5 |
| GPT-6 Sol | 13.0 |
| Claude Fable 5 | 12.5 |
| Muse Wonder Full | 12.2 |
| Claude Sonnet 5.5 | 11.9 |
| Kimi K3.1 Pro | 11.5 |
| GPT-5.6 Sol | 9.2 |
| Sonnet 5 | 7.5 |
| Claude Opus 5 | 7.3 |
| Claude Haiku 5.5 | 7.1 |
| GPT-6 Luna | 6.4 |
| Muse Spark 1.3 | 5.7 |
| MiMo-V2.6-Flash | 5.5 |
| DeepSeek V4.1 Flash 1017 | 5.2 |
| Muse Wonder | 4.4 |
| MiMo-V2.6-Pro | 2.7 |
| GPT-5.6 Terra | 2.5 |
| GLM-5.2 | 2.4 |
| Hy4 Preview | 1.7 |
| Step 5 Preview | 1.7 |
| DeepSeek V4.1 Flash | 1.5 |
| Qwen3.8-Max | 1.4 |
| Kimi K3 | 1.3 |
| GLM-5.3 | 0.6 |
| Grok 4.6 | 0.4 |
| Hy3 | 0.4 |
| Mistral Large 4 Preview | 0.3 |
| Gemini 3.8 Flash | 0.1 |
| DeepSeek V4 Pro 0813 | -0.4 |
| Grok 4.7 | -0.6 |
| Qwen3.8-Flash-Next | -1.8 |
| GLM-5.3 Flash | -3.3 |
| Qwen3.8 27B | -3.5 |
| Gemini 3.7 Flash | -6.2 |
| Gemini 3.1 Pro Preview | -8.6 |
| MiniMax M3 | -10.8 |