VLM-DeflectionBench - Realistic Accuracy: 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 accuracy in percent with gold and distractor evidence mixed; higher is better.
Metric: Accuracy (%) of knowledge-based visual question answering in the Realistic retrieval scenario (two gold and two negative contexts mixed), judged by GPT-4o with the SimpleQA protocol; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 19 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Ovis2-16B | 61.1 | #1594 |
| 2 | GPT-5 | 59.5 | #178 |
| 3 | Gemini 2.5 Pro | 51 | #276 |
| 4 | InternVL3-38B | 50.3 | #939 |
| 5 | Ovis2-34B | 49.1 | #1052 |
| 6 | Keye-VL-1.5-8B | 47.6 | #1720 |
| 7 | VL-Rethinker-72B | 47.2 | #1244 |
| 8 | Gemini 2.5 Flash | 47 | #463 |
| 9 | Qwen 2.5 VL 32B Instruct | 45.2 | #1062 |
| 10 | InternVL3-9B | 45.1 | #1728 |
| 11 | MiniCPM-V-4.5 | 44.3 | #1441 |
| 12 | Gemma 3 27B | 42.9 | #1663 |
| 13 | Pixtral-12B | 42.6 | #2553 |
| 14 | Aria | 39.5 | #2172 |
| 15 | GLM-4.1V-9B (Thinking) | 37.4 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=vlm-deflectionbench-realistic-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-21.