FACTS Leaderboard — leaderboard
FACTS grounding and quality evaluation: measures both factual grounding accuracy and response quality, combining them into a single score for open-source LLMs.
Metric: Combined Score (%). Source: huggingface.co. Status: saturation imminent. 34 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 Distill Qwen 14B | 45.76 |
| 2 | Llama 3.3 70B Instruct | 42.55 |
| 3 | Qwen 3 30B A3B | 42.55 |
| 4 | Qwen 3 4B | 42.55 |
| 5 | Qwen 3 32B | 41.7 |
| 6 | DeepSeek R1 0528 Qwen3 8B | 41.1 |
| 7 | DeepSeek R1 Distill Llama 8B | 40.68 |
| 8 | Qwen 3 8B | 40 |
| 9 | Qwen 3 14B | 38.3 |
| 10 | Gemma 3 27B (IT) | 37.8 |
| 11 | Qwen 2.5 VL 32B Instruct | 35.74 |
| 12 | Llama 3.1 70B Instruct | 33.47 |
| 13 | Gemma 3 12B (IT) | 31.3 |
| 14 | Gemma 3 4B (IT) | 30 |
| 15 | Qwen 3 1.7B | 29.79 |
Interactive version: theaggregate.ai/benchmark?slug=facts-leaderboard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.