WildTableBench - Verification: leaderboard

Accuracy in percent on the verification category: true or false claims about cell values, aggregates and filtered counts, 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 Verification question category (C3, 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 February 2027. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro71.6#145
2Seed 2.0 ProHigh68.7#222
3Kimi K2.5Thinking67.2#266
4Claude Opus 4.6High61.2#113
5Gemini 3 Flash58.2#175
6GPT-5.2High56.7#185
7Claude Sonnet 4.6High49.3#173
8Qwen 3 VL 32B Instruct41.8#606
9Qwen 3 VL 235B A22B (Thinking)41.8#459
10GPT-5 MiniHigh41.8#364
11Qwen 3 VL 32B (Thinking)40.3#573
12Qwen 3 VL 235B A22B Instruct34.3#532
13GLM-4.6V34.3#710
14O3High32.8#237
15Qwen 3 VL 8B (Thinking)28.4#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-verification · How It Works · Data refreshed daily, snapshot 2026-11-21.