FaithJudge - RAGTruth Summarization: leaderboard
Share of the model's 150 RAGTruth news summaries that the FaithJudge judge flags as hallucinated.
Metric: Hallucination rate (%) on 150 RAGTruth summaries. Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 35 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 2.5 Pro | 4.67 | #274 | |
| 2 | Gemini 2.5 Flash | 4.67 | #467 | |
| 3 | Gemini 2.0 Flash (001) | 6.67 | #727 | |
| 4 | Phi-4 | 8 | #1925 | |
| 5 | DeepSeek R1 0528 | 8 | #431 | |
| 6 | Llama 3.3 70B Instruct | 8.67 | #1387 | |
| 7 | GPT-3.5 Turbo (0125) | 8.67 | #3575 | |
| 8 | Gemini 2.5 Pro (Preview 03-25) | 9.33 | #239 | |
| 9 | glm-4-9B-chat-hf | 9.33 | #3133 | |
| 10 | GPT-4o (2024-11-20) | 10 | #864 | |
| 11 | Mistral Small 3 | 10 | #1799 | |
| 12 | GPT-4.5 (Preview) | 10 | #499 | |
| 13 | GPT-4.1 Mini | 10.67 | #793 | |
| 14 | Claude 3.7 Sonnet (20250219) | Thinking | 10.67 | #421 |
| 15 | GPT-4.1 | 11.33 | #505 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=faithjudge-ragtruth-summarization · How It Works · Data refreshed daily, snapshot 2026-11-17.