LMArena Agent Arena Work Net Improvement (Steering Burden)
Agents · 2026-08-14
Agent Arena Work category net-improvement aggregate: the overall board's five-signal net improvement recomputed on sessions tagged as office work such as document creation, professional research, planning and professional writing. A session can carry several category tags and every category session is also on the overall board, so this is a component of the overall Agent Arena family rather than a separate primary.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 20.4 |
| Claude Fable 5.1 | 14.2 |
| GPT-6 Astra | 14.1 |
| GPT-6.1 Sol | 14.0 |
| Claude Sonnet 5.5 | 12.4 |
| Gemini 4 Argon | 11.5 |
| Muse Wonder Full | 9.2 |
| GPT-6 Sol | 9.1 |
| Claude Fable 5 | 8.4 |
| Claude Opus 5 | 7.7 |
| GPT-5.6 Sol | 6.9 |
| Sonnet 5 | 6.0 |
| DeepSeek V4.1 Flash 1017 | 5.8 |
| GPT-5.5 | 4.0 |
| MiMo-V2.6-Pro | 3.9 |
| Muse Spark 1.3 | 3.7 |
| Muse Wonder | 3.5 |
| GPT-6 Luna | 3.4 |
| GLM-5.2 | 3.0 |
| MiMo-V2.6-Flash | 3.0 |
| Grok 4.7 | 3.0 |
| DeepSeek V4.1 Flash | 2.8 |
| Hy4 Preview | 2.5 |
| Gemini 3.8 Flash | 2.3 |
| Kimi K3 | 2.2 |
| Claude Haiku 5.5 | 2.1 |
| GPT-5.4 | 1.9 |
| GLM-5.3 | 1.7 |
| Qwen3.8-Max | 1.5 |
| DeepSeek V4 Pro 0813 | 0.8 |
| GPT-5.6 Terra | 0.8 |
| Grok 4.6 | 0.1 |
| Step 5 Preview | -0.4 |
| GPT-5.6 Luna | -1.5 |
| GLM-5.3 Flash | -1.8 |
| Gemini 3.7 Flash | -2.5 |
| Qwen3.8-Flash-Next | -2.8 |
| Qwen3.8 27B | -3.3 |
| Mistral Large 4 Preview | -4.5 |
| Gemini 3.1 Pro Preview | -4.7 |