RAGTruth: leaderboard
Retrieval-augmented generation hallucination corpus for evaluating grounded factuality and hallucination detection in RAG outputs.
Metric: RAG Truth (self-reported). Source: benchmarklist.com. Status: saturated. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 3.5 Sonnet | 86.1 |
| 2 | Mistral Large 2 (Jul) | 85.9 |
| 3 | GPT-4o (2024-05-13) | 84.3 |
| 4 | Llama 3.1 405B Instruct | 82.9 |
| 5 | Llama 3.3 70B Instruct | 82.6 |
| 6 | Qwen 2.5 72B Instruct | 81.9 |
| 7 | QwQ 32B-Preview | 72.4 |
Interactive version: theaggregate.ai/benchmark?slug=ragtruth · How It Works · Data refreshed daily, snapshot 2026-09-05.