FaithJudge - RAGTruth Question Answering: leaderboard

Share of the model's 139 passage-grounded answers (RAGTruth QA) that the FaithJudge judge flags as hallucinated.

Metric: Hallucination rate (%) on 139 RAGTruth QA items. Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 35 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 2.5 Pro (Preview 03-25)0.72#239
2Gemini 2.0 Flash (001)0.72#727
3Gemini 2.5 Pro1.44#274
4Gemini 2.5 Flash2.88#467
5GPT-4.1 Mini3.6#793
6Llama 3.3 70B Instruct4.32#1387
7GPT-4.14.32#505
8Phi-44.32#1925
9DeepSeek R1 05284.32#431
10GPT-4o (2024-11-20)5.04#864
11Claude Sonnet 4 (20250514)5.04#405
12Phi-4 Mini Instruct5.04#6499
13GPT-4.5 (Preview)5.04#499
14Claude Opus 4 (20250514)Thinking5.04#300
15GPT-3.5 Turbo (0125)5.76#3575

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-question-answering · How It Works · Data refreshed daily, snapshot 2026-11-17.