VisReason (Vision-Centric) - Open-Ended: leaderboard

VisReason Open-Ended scores multimodal LLMs on the 309 open-ended questions of VisReason (2026), a benchmark of vision-centric reasoning in everyday images, pooled over its 10 reasoning categories. The score is accuracy in percent, higher is better.

Metric: Accuracy (%) on the 309 open-ended questions (judged correct or not by GPT-5-mini against the reference answer) of VisReason, pooled over the 10 reasoning categories, zero-shot with chain-of-thought prompt templates, one per answer format; rows not shaded gray in the paper use explicit reasoning (GPT-5 and Gemini 3 Pro at low reasoning effort); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro (Preview)Low72.5#119
2GPT-5Low61.5#178
3GPT-5 MiniHigh60.2#364
4GPT-5.2Thinking52.1#185
5O4 Mini48.5#368
6Qwen 3 VL 235B A22B (Thinking)43.7#459
7GPT-5 Nano29.8#906
8Qwen 3 VL 32B (Thinking)29.8#573
9Qwen 3 VL 8B Instruct28.6#1029
10Qwen 3 VL 30B A3B (Thinking)25.6#719
11Qwen 3 VL 8B (Thinking)25.1#910
12Qwen 2.5 VL 32B Instruct23.9#1062
13Gemini 2.5 Flash12#463
14InternVL3-14B11.7#1214
15GPT-4o10.4#751

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-open-ended · How It Works · Data refreshed daily, snapshot 2026-11-21.