Artificial Analysis Omniscience Hallucination Rate
Safety · 2025-11-16
Artificial Analysis Omniscience Hallucination Rate is the proportion of incorrect answers among non-correct responses: incorrect / (incorrect + partial answers + not attempted). The parent evaluation contains 6,000 questions, but this metric's denominator varies with model correctness. Lower values are better.
Top models (lower is better)
| Model | Score |
|---|---|
| MiniCPM5-1B | 0.9 |
| G9v3-3B | 11.7 |
| G9v3-39A5B | 13.0 |
| Command A+ (May 2026, BF16) | 14.2 |
| Gemini 4 Argon | 15.1 |
| LFM2.5-2.6B | 16.0 |
| Grok 4.3 | 16.9 |
| Grok 4.20 | 17.4 |
| Qwen3.8 27B | 18.1 |
| MiniMax M3 | 18.4 |
| Quasar 438B | 21.4 |
| MiniCPM5-2B | 21.9 |
| Ling 3.0 Flash VL | 22.0 |
| K-EXAONE 2.0 (0803) | 22.6 |
| Muse Wonder Full | 23.0 |
| Grok 4.6 | 24.0 |
| Muse Spark Open | 24.1 |
| MiMo-V2.5-Pro | 24.7 |
| Solar Open2 250B | 25.4 |
| Granite 4.2 30B | 25.6 |
| Qwen3.7-Max (source unspecified) | 25.6 |
| Haiku 4.5 | 25.7 |
| Grok 3 Mini | 25.8 |
| K2 Horizon 375B A23B | 25.9 |
| GLM-5.2 | 26.3 |
| Granite 4.2 3B | 26.3 |
| Muse Wonder | 26.5 |
| GLM-5.3 Flash | 27.6 |
| Qwen3.7-Plus (source unspecified) | 27.7 |
| Motif 3 | 28.4 |
| Qwen3.8 Max (0902) | 28.8 |
| Sonnet 4 | 29.1 |
| Grok 4.7 | 29.3 |
| GLM-5.3 | 29.6 |
| Nemotron 3 Ultra 550B A55B | 29.7 |
| GLM-5.1 | 29.9 |
| MiMo-V2-Pro | 30.0 |
| Gemma 3 270M | 30.4 |
| Ling 3.0 Tiny | 30.5 |
| K2 Horizon MoVA 36B A4B | 30.5 |