VLM-DeflectionBench - Adversarial Hallucination Rate: leaderboard
VLM-DeflectionBench (2026) tests knowledge-based visual question answering where the right response may be to decline when retrieved evidence is missing or misleading; a GPT-4o judge grades each answer correct, deflected or hallucinated. This board is the share of hallucinated answers, wrong rather than correct or deflected, in percent with only misleading evidence; lower is better.
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: Estimated already saturated. 19 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4 | 11.1 | #332 |
| 2 | Claude Sonnet 4 | 12.1 | #414 |
| 3 | Mistral Small 3.1 | 15.6 | #1734 |
| 4 | Gemini 2.5 Pro | 22.2 | #280 |
| 5 | GLM-4.1V-9B (Thinking) | 22.5 | |
| 6 | Gemini 2.5 Flash | 24.1 | #466 |
| 7 | GPT-5 | 34.7 | #185 |
| 8 | Gemma 3 27B | 39.4 | #1672 |
| 9 | InternVL3-38B | 42.3 | #948 |
| 10 | MiniCPM-V-4.5 | 51.9 | #1451 |
| 11 | Keye-VL-1.5-8B | 53.8 | #1729 |
| 12 | VL-Rethinker-72B | 55.4 | #1252 |
| 13 | Aria | 56.4 | #2190 |
| 14 | Pixtral-12B | 56.8 | #2611 |
| 15 | Ovis2-34B | 58.1 | #1061 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=vlm-deflectionbench-adversarial-hallucination-rate · How It Works · Data refreshed daily, snapshot 2026-11-22.