WildTableBench - Color: leaderboard

Accuracy in percent on the color category: identifying cell colors, counting color-coded cells and reasoning over a color-selected subset, free-form questions about real-world table images collected from the web (WildTableBench, 2026), answers judged against the reference by GPT-5.2. Higher is better.

Metric: Accuracy (%) on the Color question category (C5, question-weighted over its subtypes), 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 October 2027. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro55.8#146
2Gemini 3 Flash37.5#180
3Kimi K2.5Thinking37.5#272
4Seed 2.0 ProHigh32.5#226
5GPT-5.2High29.2#190
6Qwen 3 VL 32B (Thinking)28.3#577
7Qwen 3 VL 235B A22B (Thinking)26.7#462
8GLM-4.6V25.8#721
9Qwen 3 VL 235B A22B Instruct21.7#541
10Claude Opus 4.6High20.8#114
11Qwen 3 VL 32B Instruct18.3#607
12GPT-5 MiniHigh18.3#373
13Claude Sonnet 4.6High17.5#179
14O3High15#242
15Qwen 3 VL 8B (Thinking)11.7#924

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=wildtablebench-color · How It Works · Data refreshed daily, snapshot 2026-11-22.