OpenAI's factuality benchmark — 4,326 short-answer factual questions with verified, unambiguous answers. Specifically designed to measure hallucination rates.
Directly targets factual accuracy and hallucination. All answers are independently verified. Clear right/wrong scoring.
Only tests factual recall, not reasoning. English-centric. Knowledge cutoff affects scores for time-sensitive questions.
Higher is better. Accuracy percentage (0–100%). Top models score ~70%.
Only tests factual recall, not reasoning.
| # | Model | Score |
|---|---|---|
| 1 | 96.5 | |
| 2 | 93.0 | |
| 3 | 90.6 | |
| 4 | 79.8 | |
| 5 | 75.9 | |
| 6 | 72.1 | |
| 7 | 72.1 | |
| 8 | 72.1 | |
| 9 | 72.0 | |
| 10 | 72.0 | |
| 11 | 72.0 | |
| 12 | 71.5 | |
| 13 | 70.8 | |
| 14 | 69.4 | |
| 15 | 68.9 | |
| 16 | 68.5 | |
| 17 | 68.2 | |
| 18 | 68.0 | |
| 19 | 68.0 | |
| 20 | 67.6 | |
| 21 | 66.9 | |
| 22 | 66.8 | |
| 23 | 65.9 | |
| 24 | 65.8 | |
| 25 | 65.5 | |
| 26 | 64.9 | |
| 27 | 64.3 | |
| 28 | 63.8 | |
| 29 | 63.5 | |
| 30 | 63.4 | |
| 31 | 62.8 | |
| 32 | 62.8 | |
| 33 | 62.8 | |
| 34 | 62.8 | |
| 35 | 62.5 | |
| 36 | 61.9 | |
| 37 | 61.4 | |
| 38 | 61.0 | |
| 39 | 60.8 | |
| 40 | 60.7 | |
| 41 | 60.0 | |
| 42 | 59.8 | |
| 43 | 59.3 | |
| 44 | 58.5 | |
| 45 | 58.2 | |
| 46 | 58.1 | |
| 47 | 58.0 | |
| 48 | 58.0 | |
| 49 | 58.0 | |
| 50 | 58.0 |