VLM-DeflectionBench - Parametric Hallucination Rate: leaderboard
Metric: Hallucination rate (%): share of answers that are wrong rather than correct or a deflection, in the Parametric retrieval scenario (no retrieved context), judged by GPT-4o with the SimpleQA protocol; lower is better. Source: arxiv.org. Saturation forecast: Around 2030. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Mistral Small 3.1 | 26.6 |
| 2 | Claude Opus 4 | 28.3 |
| 3 | Claude Sonnet 4 | 58.5 |
| 4 | Gemini 2.5 Pro | 65.5 |
| 5 | Gemini 2.5 Flash | 70.8 |
| 6 | GPT-5 | 74.8 |
| 7 | Pixtral-12B | 76 |
| 8 | GLM-4.1V-9B (Thinking) | 87.4 |
Interactive version: theaggregate.ai/benchmark?slug=vlm-deflectionbench-parametric-hallucination-rate · How It Works · Data refreshed daily, snapshot 2026-10-07.