Delulu - Parameter Hallucination Detection: leaderboard
Metric: Both-correct rate (%) on the 435 Delulu samples whose hallucination is parameters that do not exist, 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: Around January 2027. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 91.1 |
| 2 | Claude Sonnet 4.6 | 88.5 |
| 3 | GPT-5.2 Codex | 87.2 |
| 4 | GPT-5.4 | 79.8 |
| 5 | Claude Haiku 4.5 | 74.3 |
| 6 | GPT-5.4 Mini | 66.3 |
| 7 | GPT-4.1 Mini | 63.5 |
| 8 | GPT-4o Mini | 46.1 |
Interactive version: theaggregate.ai/benchmark?slug=delulu-parameter-hallucination-detection · How It Works · Data refreshed daily, snapshot 2026-10-07.