MMGist - Visual Logic: leaderboard

Metric: Accuracy (%) on the Visual Logic dimension (items kept from LogicVista and ZeroBench); MMGist: 7,262 curated items from 18 vision-language benchmarks after removing items answerable without the image, items nearly every model solves and items with faulty labels; eight samples per item at temperature 1.0, step-by-step answer in a boxed block, 16,384 output tokens, medium effort where configurable; higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 27 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview) (Medium)40.4
2Seed 2.0 Pro37.7
3Seed 2.0 Lite35.1
4Qwen 3.6 Plus34.9
5Qwen 3.6 35B A3B31.5
6Seed 2.0 Mini30.3
7Qwen 3.5 27B28.1
8Gemma 4 31B27.7
9Gemini 3.1 Flash Lite (Medium)27.2
10Qwen 3.5 9B26.4
11GPT-5 (Medium)25.8
12GPT-5 Mini (Medium)24.4
13Gemma 4 26B A4B24.3
14Claude Sonnet 4.6 (Medium)22.4
15Step3 VL 10B21.6

Interactive version: theaggregate.ai/benchmark?slug=mmgist-visual-logic · How It Works · Data refreshed daily, snapshot 2026-09-29.