FaithJudge - FaithBench Summarization: leaderboard

Share of the model's 72 FaithBench article summaries that the FaithJudge judge flags as hallucinated.

Metric: Hallucination rate (%) on 72 FaithBench summaries. Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 35 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 2.5 Flash19.44#467
2DeepSeek R1 052820.83#431
3Gemini 2.5 Pro25#274
4Gemini 2.5 Pro (Preview 03-25)25#239
5Gemini 2.0 Flash (001)29.17#727
6GPT-4.136.11#505
7Qwen 2.5 32B Instruct36.11#1257
8O3 (2025-04-16)Medium36.11#229
9GPT-4.1 Mini37.5#793
10GPT-4.5 (Preview)37.5#499
11Claude 3.7 Sonnet (20250219)38.89#421
12Jamba 1.6 Mini38.89#5461
13Claude Opus 4 (20250514)Thinking38.89#300
14glm-4-9B-chat-hf38.89#3133
15GPT-4o (2024-11-20)40.28#864

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=faithjudge-faithbench-summarization · How It Works · Data refreshed daily, snapshot 2026-11-17.