WildTableBench - Color Identification: leaderboard

A WildTableBench (2026) subtype on naming the color of a cell or region, or the category a color encodes, 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 Identification subtype (C5-I), 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: Not forecast. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 VL 235B A22B Instruct66.7#532
2GLM-4.6V66.7#710
3Kimi K2.5Thinking66.7#266
4Gemini 3 Pro55.6#145
5Qwen 3 VL 235B A22B (Thinking)55.6#459
6Qwen 3 VL 32B (Thinking)55.6#573
7Seed 2.0 ProHigh55.6#222
8Gemini 3 Flash50#175
9GPT-5.2High50#185
10Qwen 3 VL 8B Instruct44.4#1029
11Qwen 3 VL 32B Instruct44.4#606
12GPT-5 MiniHigh44.4#364
13Claude Opus 4.6High44.4#113
14Qwen 3 VL 4B Instruct38.9#1332
15Qwen 3 VL 2B Instruct38.9#2367

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