VisReason (Vision-Centric) - Pattern Counting: leaderboard

VisReason Pattern Counting is a perceptual-reasoning category of VisReason (2026): the model must identify all target patterns in a complex scene and count their occurrences. The score is accuracy in percent, higher is better.

Metric: Accuracy (%) on the Pattern Counting category (perceptual reasoning: identify all target patterns in a complex scene and count their occurrences) of VisReason, answers scored by format (regular-expression match, GPT-5-mini judge for open-ended answers, IoU above 0.5 for boxes), 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. 20 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro (Preview)Low50#119
2GPT-5 MiniHigh37#364
3GPT-5.2Thinking26.1#185
4O4 Mini21.7#368
5GPT-5Low21.7#178
6Qwen 3 VL 235B A22B (Thinking)19.6#459
7GPT-5 Nano17.4#906
8Qwen 3 VL 32B (Thinking)17.4#573
9Qwen 3 VL 30B A3B (Thinking)15.2#719
10Keye-VL-1.5-8BThinking10.9#1720
11MiMo-VL-7B-RL10.6#1155
12GPT-4o8.7#751
13Gemini 2.5 Flash8.7#463
14Qwen 3 VL 8B Instruct8.6#1029
15InternVL3-14B6.5#1214

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-pattern-counting · How It Works · Data refreshed daily, snapshot 2026-11-21.