FaithJudge - RAGTruth Data-to-Text: leaderboard

Share of the model's 150 business overviews written from structured JSON (RAGTruth data-to-text) that the FaithJudge judge flags as hallucinated.

Metric: Hallucination rate (%) on 150 RAGTruth data-to-text 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)4#239
2Gemini 2.5 Pro4.67#274
3Gemini 2.5 Flash4.67#467
4GPT-4.18#505
5GPT-4.5 (Preview)8#499
6DeepSeek R1 052811.33#431
7Claude 3.7 Sonnet (20250219)12.67#421
8Gemini 2.0 Flash (001)13.33#727
9Claude Opus 4 (20250514)Thinking13.33#300
10Grok 314#606
11GPT-3.5 Turbo (0125)15.33#3575
12O3 (2025-04-16)Medium15.33#229
13Mistral Small 318#1799
14Claude Sonnet 4 (20250514)Thinking19.33#405
15Qwen 2.5 72B Instruct20#1119

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-data-to-text · How It Works · Data refreshed daily, snapshot 2026-11-17.