FACTS Benchmark Suite
General QA · 2025-12-11
The FACTS Benchmark Suite is Google DeepMind's aggregate factuality score, the unweighted average of four component leaderboards: FACTS Grounding, FACTS Parametric, FACTS Search, and FACTS Multimodal. It is a later and broader construct than FACTS Grounding alone, which it contains as one component, and the two are not interchangeable.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 75.7 |
| GPT-6.1 Sol | 75.3 |
| GPT-6 Sol | 72.5 |
| Gemini 3.7 Flash | 71.2 |
| Gemini 3 Pro Preview | 70.5 |
| Gemini 3.8 Flash | 69.4 |
| Gemini 3.1 Pro Preview | 67.4 |
| Gemini 3.6 Flash | 65.3 |
| Claude Opus 5.5 | 65.3 |
| Gemini 2.5 Pro | 63.4 |
| Gemini 3 Flash Preview | 63.4 |
| GPT-6 Luna | 63.3 |
| Gemini 3.5 Flash | 63.1 |
| GPT-5.2 | 61.4 |
| Opus 4.8 | 53.5 |
| Gemini 3.5 Flash-Lite | 53.5 |
| GPT-5.6 Terra | 52.0 |
| GPT-5.1 | 50.8 |
| Gemini 2.5 Flash | 50.4 |
| Grok 4.5 | 50.2 |
| Gemma 4 31B IT | 45.3 |
| Gemma 4 26B A4B IT | 42.7 |
| Grok 4.1 Fast | 42.1 |
| Gemini 3.1 Flash-Lite Preview | 40.6 |
| GPT-5.6 Luna | 35.0 |
| GPT-5 Mini | 33.7 |
| Haiku 4.5 | 18.6 |
| Gemini 2.5 Flash-Lite | 17.9 |