VLM-DeflectionBench - Realistic Accuracy: leaderboard
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 2029. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 59.5 |
| 2 | Gemini 2.5 Pro | 51 |
| 3 | InternVL3-38B | 50.3 |
| 4 | Gemini 2.5 Flash | 47 |
| 5 | Qwen 2.5 VL 32B Instruct | 45.2 |
| 6 | Gemma 3 27B | 42.9 |
| 7 | Pixtral-12B | 42.6 |
| 8 | GLM-4.1V-9B (Thinking) | 37.4 |
| 9 | Claude Sonnet 4 | 33.3 |
| 10 | Claude Opus 4 | 32.1 |
| 11 | Mistral Small 3.1 | 23.5 |
Interactive version: theaggregate.ai/benchmark?slug=vlm-deflectionbench-realistic-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-07.