LMArena Agent Arena Chat Net Improvement (Steering Burden)
Agents · 2026-08-14
Agent Arena Chat category net-improvement aggregate: the overall board's five-signal net improvement recomputed on sessions tagged as general agentic chat such as creative writing, learning, everyday research and media generation. 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 | 23.3 |
| Claude Sonnet 5.5 | 17.1 |
| Claude Fable 5.1 | 15.2 |
| Gemini 4 Argon | 14.8 |
| GPT-6 Astra | 11.7 |
| GPT-6.1 Sol | 11.6 |
| Muse Wonder Full | 10.8 |
| Claude Opus 5 | 9.6 |
| Gemini 3.8 Flash | 8.2 |
| Kimi K3.1 Pro | 7.7 |
| Claude Fable 5 | 6.7 |
| GPT-5.6 Sol | 6.1 |
| GPT-6 Sol | 5.9 |
| Grok 4.7 | 4.8 |
| MiMo-V2.6-Pro | 4.0 |
| Gemini 3.7 Flash | 3.1 |
| Muse Spark 1.3 | 2.6 |
| DeepSeek V4.1 Flash 1017 | 2.5 |
| Hy4 Preview | 2.2 |
| Muse Wonder | 1.9 |
| DeepSeek V4.1 Flash | 1.6 |
| Kimi K3 | 1.4 |
| GPT-6 Luna | 1.2 |
| GLM-5.3 | 0.4 |
| GPT-5.6 Terra | 0.3 |
| Qwen3.8-Max | 0.3 |
| Grok 4.6 | 0.3 |
| MiMo-V2.6-Flash | 0.2 |
| Sonnet 5 | 0.0 |
| GLM-5.3 Flash | -0.3 |
| GLM-5.2 | -1.0 |
| DeepSeek V4 Pro 0813 | -1.7 |
| Gemini 3.1 Pro Preview | -2.0 |
| Muse Spark 1.1 | -2.6 |
| Claude Haiku 5.5 | -3.3 |
| Mistral Large 4 Preview | -3.9 |
| Muse Spark 1.2 | -4.7 |
| Qwen3.8-Flash-Next | -4.9 |
| Step 5 Preview | -5.0 |
| Qwen3.7-Plus (source unspecified) | -5.8 |