VisReason (Vision-Centric) - Fill-in-the-Blank: leaderboard

VisReason Fill-in-the-Blank scores multimodal LLMs on the 441 fill-in-the-blank 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 441 fill-in-the-blank questions (answer extracted by regular expression) 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)Low47.4#119
2GPT-5.2Thinking41.8#185
3GPT-5Low35.8#178
4GPT-5 MiniHigh32.4#364
5O4 Mini30.4#368
6Qwen 3 VL 32B (Thinking)22.2#573
7Qwen 3 VL 235B A22B (Thinking)21.1#459
8Qwen 3 VL 8B Instruct20.8#1029
9Qwen 3 VL 30B A3B (Thinking)20.5#719
10GPT-5 Nano20.2#906
11GPT-4o18.6#751
12Qwen 3 VL 8B (Thinking)17.5#910
13Gemini 2.5 Flash16.3#463
14Qwen 2.5 VL 32B Instruct14.5#1062
15InternVL3-8B10.2#1782

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-fill-in-the-blank · How It Works · Data refreshed daily, snapshot 2026-11-21.