Vectara Hallucination Leaderboard: leaderboard
Measures how much LLMs hallucinate when summarizing text. Uses the Hughes Hallucination Evaluation Model (HHEM) for automated detection.
Metric: Factual Consistency Rate (%). Source: github.com. 109 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 Nano (2026-03-17) | 96.9 | #266 |
| 2 | Gemini 2.5 Flash Lite | 96.7 | #422 |
| 3 | Phi-4 | 96.3 | #706 |
| 4 | Llama 3.3 70B Instruct | 95.9 | #534 |
| 5 | Gemma 3 12B (IT) | 95.6 | #664 |
| 6 | Mistral Large 2 (Nov) Instruct (2411) | 95.5 | #465 |
| 7 | Qwen 3 8B | 95.2 | #674 |
| 8 | Mistral Small 3 | 94.9 | #627 |
| 9 | Nova Pro (v1) | 94.9 | #497 |
| 10 | Nova 2.0 Lite (Non-reasoning) | 94.9 | #417 (Nova 2.0 Lite) |
| 11 | Gemma 4 26B A4B (IT) | 94.8 | #297 |
| 12 | Granite 4.0 H Small | 94.8 | #665 |
| 13 | DeepSeek V3.2 Exp | 94.7 | #234 |
| 14 | Qwen 3 14B | 94.6 | #537 |
| 15 | DeepSeek V3.1 | 94.5 | #264 |
Interactive version: theaggregate.ai/benchmark?slug=vectara-hallucination-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.