WildTableBench - Semantic Lookup and Structure: leaderboard

A WildTableBench (2026) subtype on finding the cell at a row and column label intersection or answering a structural count, 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 Semantic Lookup and Structure subtype (C1-S), 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
1Kimi K2.5Thinking56.7#272
2Seed 2.0 ProHigh56.7#226
3Gemini 3 Flash50#180
4Gemini 3 Pro46.7#146
5Qwen 3 VL 32B (Thinking)46.7#577
6GPT-5.2High46.7#190
7Claude Sonnet 4.6High46.7#179
8Qwen 3 VL 235B A22B (Thinking)33.3#462
9Claude Opus 4.6High33.3#114
10Qwen 3 VL 235B A22B Instruct23.3#541
11Qwen 3 VL 32B Instruct23.3#607
12GPT-5 MiniHigh20#373
13O3High20#242
14GLM-4.6V10#721
15Qwen 3 VL 2B Instruct6.7#2389

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

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