VLM-DeflectionBench - Adversarial Hallucination Rate: leaderboard
Metric: Hallucination rate (%): share of answers that are wrong rather than correct or a deflection, in the Adversarial retrieval scenario (negative contexts only), judged by GPT-4o with the SimpleQA protocol; lower is better. Source: arxiv.org. Saturation forecast: Around January 2027. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4 | 11.1 |
| 2 | Claude Sonnet 4 | 12.1 |
| 3 | Mistral Small 3.1 | 15.6 |
| 4 | Gemini 2.5 Pro | 22.2 |
| 5 | GLM-4.1V-9B (Thinking) | 22.5 |
| 6 | Gemini 2.5 Flash | 24.1 |
| 7 | GPT-5 | 34.7 |
| 8 | Gemma 3 27B | 39.4 |
| 9 | InternVL3-38B | 42.3 |
| 10 | Pixtral-12B | 56.8 |
| 11 | Qwen 2.5 VL 32B Instruct | 83.9 |
Interactive version: theaggregate.ai/benchmark?slug=vlm-deflectionbench-adversarial-hallucination-rate · How It Works · Data refreshed daily, snapshot 2026-10-07.