LMArena Agent Arena Net Improvement (Steering Burden)
Agents
Agent Arena net-improvement aggregate using the five September 30 signals: confirmed success, praise versus complaint, steering_burden, bash recovery and tool hallucination. The steering_burden source signal replaces steerability; preserved separately from the earlier aggregate because the source key and observation population changed.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 14.6 |
| GPT-6 Astra | 12.2 |
| GPT-6 Sol | 10.6 |
| Claude Opus 5 | 8.8 |
| Claude Fable 5 | 8.4 |
| Gemini 4 Argon | 7.9 |
| GPT-5.6 Sol | 7.1 |
| Opus 4.8 | 6.9 |
| Sonnet 5 | 4.8 |
| Hy4 Preview | 4.2 |
| Muse Spark 1.3 | 4.2 |
| GPT-5.5 | 4.1 |
| DeepSeek V4.1 Flash | 4.0 |
| Kimi K3 | 4.0 |
| Grok 4.7 | 3.6 |
| GLM-5.2 | 3.6 |
| Gemini 3.8 Flash | 3.0 |
| Qwen3.8-Max | 2.8 |
| GLM-5.3 | 2.6 |
| DeepSeek V4 Pro 0813 | 1.3 |
| Grok 4.6 | 1.2 |
| Grok 4.5 | 1.1 |
| GPT-5.4 | 0.9 |
| GPT-5.6 Terra | 0.8 |
| GLM-5.3 Flash | -0.3 |
| Qwen3.8-Flash-Next | -0.8 |
| GPT-5.6 Luna | -1.1 |
| Qwen3.8 27B | -1.6 |
| Gemini 3.7 Flash | -1.6 |
| Muse Spark 1.2 | -3.0 |
| Muse Spark 1.1 | -4.6 |
| Qwen3.7-Max | -5.2 |
| Hy3 | -5.7 |
| MiniMax M3 | -6.7 |
| MiMo-V2.5-Pro | -7.2 |
| Qwen3.7-Plus | -7.3 |
| Gemini 3.1 Pro Preview | -7.3 |
| Gemini 3.6 Flash | -7.7 |
| Inkling-Small | -9.8 |
| Inkling | -10.9 |