LMArena Agent Arena Tool Hallucination
Agents · 2026-06-04
Signed percentage-point Agent Arena diagnostic signal for apparent nonexistent or inappropriate tool use. Values may be negative when hallucination behavior is below the leaderboard baseline; lower values indicate safer and more grounded tool use.
Top models (lower is better)
| Model | Score |
|---|---|
| Gemma 4 31B IT | -28.7 |
| Opus 4.8 | -21.9 |
| Gemini 3.5 Flash | -1.9 |
| DeepSeek-V4-Flash | -0.9 |
| Nemotron 3 Ultra 550B A55B | 0.2 |
| Qwen3.7-Plus | 0.2 |
| Gemini 3 Flash Preview | 0.2 |
| MiMo-V2.5-Pro | 0.3 |
| Hy3 | 0.3 |
| Inkling | 0.4 |
| Gemini 3.5 Flash-Lite | 0.4 |
| GLM-5.1 | 0.5 |
| DeepSeek-V4-Pro | 0.7 |
| MiniMax M3 | 0.8 |
| Qwen3.7-Max | 0.8 |
| Grok Build 0.1 | 0.9 |
| MiniMax M2.7 | 1.1 |
| Grok 4.3 | 1.1 |
| Sonnet 5 | 1.2 |
| Opus 4.7 | 1.2 |
| Sonnet 4.6 | 1.2 |
| Gemini 3.1 Pro Preview | 1.3 |
| Claude Opus 5 | 1.3 |
| Muse Spark 1.1 | 1.3 |
| Opus 4.6 | 1.3 |
| GPT-5.5 | 1.3 |
| Kimi K2.6 | 1.3 |
| GPT-5.4 | 1.3 |
| Gemini 3.6 Flash | 1.3 |
| GPT-5.6 Terra | 1.3 |
| Kimi K2.7 Code | 1.3 |
| Grok 4.5 | 1.3 |
| GPT-5.6 Sol | 1.3 |
| Claude Fable 5 | 1.3 |
| GPT-5.6 Luna | 1.3 |
| GLM-5.2 | 1.3 |
| Kimi K3 | 1.3 |