AA-Omniscience Business Non-Hallucination
Safety
AA-Omniscience's business-domain non-hallucination rate, as published in the Artificial Analysis Finance & Accounting Index: how often a model avoids fabricating an answer when it does not answer correctly. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Carbon | 92.9 |
| Muse Wonder Full | 89.3 |
| Muse Wonder | 87.0 |
| Grok 4.7 | 85.0 |
| K2 Horizon 375B A23B | 83.0 |
| Mistral Large 4 | 82.2 |
| Qwen3.8 Max (0902) | 81.4 |
| Nemotron 3.5 Super VL 120B A12B (precision unspecified) | 81.2 |
| Muse Spark Open | 81.0 |
| GPT-6.1 Terra | 79.5 |
| Qwen3.8 27B | 76.2 |
| GLM-5.3 | 75.9 |
| Claude Haiku 5.5 | 74.6 |
| Ling 3.1 Flash 1019 | 73.7 |
| Kimi K3.1 Pro | 72.1 |
| GPT-6.1 Sol | 70.2 |
| Beam 1.0 501B A23B | 70.1 |
| Step 5 Preview | 68.4 |
| GPT-6 Astra | 67.8 |
| Claude Sonnet 5.5 | 62.9 |
| Gemini 3.8 Flash | 59.7 |
| MiMo-V2.6-Pro | 48.4 |
| DeepSeek V4.1 Flash 1017 | 44.7 |
| Inkling | 39.1 |
| GPT-6 Luna | 28.0 |
| MiniMax M3.1 Flash | 10.2 |