Delulu - Undefined Variable Hallucination Detection: leaderboard

LLMs as judges on the 577 Delulu (2026) fill-in-the-middle samples whose hallucinated completion contains references to undefined variables: the judge must accept the real completion and reject the hallucinated one. The score is the both-correct rate (%); higher is better.

Metric: Both-correct rate (%) on the 577 Delulu samples whose hallucination is references to undefined variables, 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 4.598.1#160
2GPT-5.2 Codex96.2#182
3GPT-5.494.6#135
4Claude Sonnet 4.694.3#173
5Claude Haiku 4.591.5#583
6GPT-5.4 Mini83.1#406
7GPT-4.1 Mini75.5#789
8GPT-4o Mini58#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-undefined-variable-hallucination-detection · How It Works · Data refreshed daily, snapshot 2026-11-21.