Delulu - Hallucination Detection: leaderboard

LLMs as judges on Delulu (2026): each of 1,951 fill-in-the-middle samples is shown once with the real completion and once with an execution-verified hallucinated one, and the judge must accept the first and reject the second. The score is the both-correct rate (%); higher is better.

Metric: Both-correct rate (%) on all 1,951 Delulu samples as an LLM judge: the judge sees each sample twice, once with the gold completion and once with the hallucinated one, and must accept the first and reject the second (both-correct rate, %), with chain-of-thought and a binary verdict; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.592.1#160
2GPT-5.2 Codex88.2#182
3Claude Sonnet 4.685.8#173
4GPT-5.482.8#135
5Claude Haiku 4.577.9#583
6GPT-5.4 Mini74.8#406
7GPT-4.1 Mini67.7#789
8GPT-4o Mini52.3#1667

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

Interactive version: theaggregate.ai/benchmark?slug=delulu-hallucination-detection · How It Works · Data refreshed daily, snapshot 2026-11-21.