WildTableBench - Value Verification: leaderboard

A WildTableBench (2026) subtype on judging a factual claim about specific cell values true or false, 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 Value Verification subtype (C3-V), 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 Pro73.9#145
2Claude Opus 4.6High73.9#113
3Seed 2.0 ProHigh73.9#222
4Kimi K2.5Thinking56.5#266
5GPT-5.2High52.2#185
6GPT-5 MiniHigh52.2#364
7Claude Sonnet 4.6High52.2#173
8O3High47.8#237
9Gemini 3 Flash43.5#175
10Qwen 3 VL 32B Instruct43.5#606
11Qwen 3 VL 32B (Thinking)39.1#573
12Qwen 3 VL 8B (Thinking)34.8#910
13Qwen 3 VL 235B A22B Instruct30.4#532
14Qwen 3 VL 4B Instruct26.1#1332
15Qwen 3 VL 235B A22B (Thinking)26.1#459

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

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