VLM-DeflectionBench - Oracle 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 only gold evidence; higher is better.

Metric: Accuracy (%) of knowledge-based visual question answering in the Oracle retrieval scenario (gold evidence only), judged by GPT-4o with the SimpleQA protocol; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-573.1#185
2Ovis2-34B66.5#1061
3MiniCPM-V-4.565.3#1451
4Ovis2-16B64.4#1608
5InternVL3-38B64.2#948
6Keye-VL-1.5-8B63.4#1729
7Pixtral-12B62.7#2611
8VL-Rethinker-72B62.6#1252
9Aria62.3#2190
10InternVL3-9B61.2#1746
11Qwen 2.5 VL 32B Instruct61#1065
12Gemini 2.5 Pro59.8#280
13Gemma 3 27B59.5#1672
14Gemini 2.5 Flash58.8#466
15GLM-4.1V-9B (Thinking)55.2

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-oracle-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-22.