AA-Omniscience v1.1 Hallucination Rate
Safety · 2026-10-27
AA-Omniscience v1.1 Hallucination Rate is the proportion of incorrect answers among non-correct responses: incorrect / (incorrect + partial answers + not attempted). The v1.1 evaluation prompts 3,000 questions, but this metric's denominator varies with model correctness. Lower values are better.
Top models (lower is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 4.3 |
| Command A+ (May 2026, BF16) | 13.9 |
| Muse Wonder Full | 17.6 |
| Qwen3.8 27B | 18.0 |
| Grok 4.7 | 20.0 |
| Muse Wonder | 20.9 |
| Muse Spark 1.3 | 22.5 |
| Nemotron 3 Ultra 550B A55B | 22.6 |
| Gemini 3.5 Flash-Lite | 26.7 |
| Qwen3.8 Max (0902) | 27.1 |
| Muse Spark Open | 28.0 |
| K2 Horizon 375B A23B | 29.0 |
| Nemotron 3.5 Super VL 120B A12B (precision unspecified) | 30.5 |
| Step 5 Preview | 31.3 |
| Claude Haiku 5.5 | 34.2 |
| Qwen3.8 2.4T A95B | 34.5 |
| GPT-6.1 Terra | 34.5 |
| Ling 3.1 Flash 1019 | 36.3 |
| Solar Mini 4 | 36.5 |
| Kimi K3.1 Pro | 37.3 |
| GLM-5.3 Flash | 38.3 |
| GLM-5.3 | 38.9 |
| Beam 1.0 501B A23B | 41.0 |
| Mistral Large 4 Preview | 41.3 |
| GPT-6 Astra | 42.5 |
| Claude Sonnet 5.5 | 43.3 |
| Qwen3.8-Flash-Next | 46.8 |
| GPT-6.1 Sol | 48.0 |
| Gemini 3.8 Flash | 51.5 |
| MiMo-V2.6-Flash | 52.2 |
| MiMo-V2.6-Pro | 60.2 |
| Inkling-Small | 60.5 |
| DeepSeek V4.1 Flash 1017 | 64.1 |
| Mistral Small 4 | 66.6 |
| Inkling | 67.2 |
| Kimi K3 | 76.6 |
| GPT-6 Luna | 79.3 |
| Mistral Medium 3.5 | 80.2 |
| Muse Glimmer 30B | 83.6 |
| Ring 2.6 1T | 84.6 |