Delulu is a multi-lingual benchmark showing that top code LLMs still hallucinate in FIM tasks, with the strongest model reaching only 84.5% pass@1.
arXiv.org , year =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Delulu: A Verified Multi-Lingual Benchmark for Code Hallucination Detection in Fill-in-the-Middle Tasks
Delulu is a multi-lingual benchmark showing that top code LLMs still hallucinate in FIM tasks, with the strongest model reaching only 84.5% pass@1.