WildTableBench - Color-based Reasoning: leaderboard

A WildTableBench (2026) subtype on counting or computing over the cells selected by a color filter plus another condition, asked about real-world table images from the web. Accuracy in percent, free-form answers judged against the reference by GPT-5.2; higher is better.

Metric: Accuracy (%) on the Color-based Reasoning subtype (C5-R), 928 human-verified questions over 402 real-world table images, free-form answers judged by GPT-5.2 against the reference answer, high reasoning effort where configurable; higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro51.6#145
2Kimi K2.5Thinking35.5#266
3Gemini 3 Flash33.9#175
4Qwen 3 VL 32B (Thinking)22.6#573
5GPT-5.2High22.6#185
6Seed 2.0 ProHigh22.6#222
7Qwen 3 VL 235B A22B (Thinking)21#459
8GLM-4.6V19.4#710
9O3High14.5#237
10Qwen 3 VL 235B A22B Instruct12.9#532
11GPT-5 MiniHigh12.9#364
12Claude Sonnet 4.6High12.9#173
13Claude Opus 4.6High11.3#113
14Qwen 3 VL 32B Instruct9.7#606
15Qwen 3 VL 8B (Thinking)6.5#910

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

Interactive version: theaggregate.ai/benchmark?slug=wildtablebench-color-based-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.