LongFact-Objects hallucination rate - OpenAI grader
General QA · 2025-08-07
Share of factually incorrect claims in long-form answers to LongFact-Objects prompts, fact-checked by OpenAI's LLM grader with browsing; the evaluated model has no tools. OpenAI-specific grading pipeline; diagnostic. Lower is better.
Top models (lower is better)
| Model | Score |
|---|---|
| GPT-4.1 | 1.1 |
| GPT-5 Mini | 1.3 |
| GPT-4.1 Mini | 1.8 |
| GPT-5 Nano | 2.8 |
| O4 Mini | 8.9 |