Factual consistency rate from the Vectara/Hughes Hallucination Evaluation — measures how often model-generated summaries remain faithful to source documents without fabricating information.
Tests generation faithfulness rather than knowledge recall. Large model coverage (95+ models). Simple, reproducible methodology based on document summarization.
Only tests summarization faithfulness, not open-ended generation. Models can score well by being overly extractive. Answer rate varies across models.
Higher is better. Percentage (0-100) of generated summaries that are factually consistent with the source document.
Max score: 100
Only tests summarization faithfulness, not open-ended generation.
| # | Model | Score |
|---|---|---|
| 1 | 96.9 | |
| 2 | 96.9 | |
| 3 | 96.9 | |
| 4 | 96.7 | |
| 5 | 96.7 | |
| 6 | 96.6 | |
| 7 | 96.3 | |
| 8 | 96.2 | |
| 9 | 96.0 | |
| 10 | 95.8 | |
| 11 | 95.8 | |
| 12 | 95.7 | |
| 13 | 95.7 | |
| 14 | 95.6 | |
| 15 | 95.6 | |
| 16 | 95.5 | |
| 17 | 95.2 | |
| 18 | 95.2 | |
| 19 | 95.2 | |
| 20 | 95.2 | |
| 21 | 95.2 | |
| 22 | 94.9 | |
| 23 | 94.9 | |
| 24 | 94.8 | |
| 25 | 94.8 | |
| 26 | 94.8 | |
| 27 | 94.8 | |
| 28 | 94.8 | |
| 29 | 94.7 | |
| 30 | 94.7 | |
| 31 | 94.7 | |
| 32 | 94.6 | |
| 33 | 94.6 | |
| 34 | 94.6 | |
| 35 | 94.6 | |
| 36 | 94.6 | |
| 37 | Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning) source benchpress Evidence observed Apr 13, 2026 | 94.6 |
| 38 | 94.5 | |
| 39 | 94.5 | |
| 40 | 94.5 | |
| 41 | 94.5 | |
| 42 | 94.5 | |
| 43 | 94.5 | |
| 44 | 94.4 | |
| 45 | 94.4 | |
| 46 | 94.3 | |
| 47 | 94.3 | |
| 48 | 94.3 | |
| 49 | 94.3 | |
| 50 | 94.2 |