LMArena Agent Arena Code Tool Hallucination
Agents · 2026-08-14
Signed percentage-point Agent Arena Code diagnostic improvement signal for reductions in apparent nonexistent or inappropriate tool use, recomputed on sessions tagged as writing or debugging code, automating workflows or analyzing data. Atlas stores the raw improvement orientation relative to the category baseline: positive values indicate more grounded tool use and higher is better. Reported confidence intervals are not standard errors.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K3.1 Pro | 0.7 |
| GPT-6.1 Sol | 0.7 |
| GLM-5.2 | 0.7 |
| Kimi K3 | 0.7 |
| GLM-5.3 Flash | 0.7 |
| Grok 4.7 | 0.7 |
| GPT-6 Astra | 0.7 |
| Claude Opus 5 | 0.7 |
| Claude Fable 5 | 0.7 |
| GLM-5.3 | 0.7 |
| Claude Fable 5.1 | 0.7 |
| Muse Spark 1.3 | 0.7 |
| GPT-6 Sol | 0.7 |
| GPT-5.6 Sol | 0.6 |
| Grok 4.6 | 0.6 |
| Step 5 Preview | 0.6 |
| GPT-6 Luna | 0.6 |
| Gemini 3.8 Flash | 0.6 |
| DeepSeek V4 Pro 0813 | 0.6 |
| DeepSeek V4.1 Flash | 0.5 |
| Claude Sonnet 5.5 | 0.5 |
| GPT-5.6 Terra | 0.5 |
| Gemini 3.7 Flash | 0.5 |
| Claude Opus 5.5 | 0.5 |
| Muse Wonder Full | 0.3 |
| DeepSeek V4.1 Flash 1017 | 0.3 |
| Gemini 4 Argon | 0.3 |
| Muse Spark 1.2 | 0.3 |
| Muse Spark 1.1 | 0.3 |
| Sonnet 5 | 0.2 |
| Qwen3.8 27B | 0.1 |
| Hy4 Preview | 0.1 |
| Muse Wonder | -0.0 |
| MiMo-V2.6-Pro | -0.1 |
| Qwen3.7-Max (source unspecified) | -0.2 |
| Claude Haiku 5.5 | -0.6 |
| Gemini 3.1 Pro Preview | -0.6 |
| MiniMax M3 | -0.6 |
| Qwen3.8-Max | -0.7 |
| Qwen3.8-Flash-Next | -1.0 |