WildTableBench - Cell Locating: leaderboard

A WildTableBench (2026) subtype on reading the value at a given cell address such as G11, 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 Cell Locating subtype (C1-L), 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 Pro71.4#145
2Seed 2.0 ProHigh67.3#222
3GPT-5.2High65.3#185
4Claude Opus 4.6High65.3#113
5Gemini 3 Flash61.2#175
6Kimi K2.5Thinking51#266
7Qwen 3 VL 235B A22B (Thinking)46.9#459
8Claude Sonnet 4.6High40.8#173
9Qwen 3 VL 32B Instruct30.6#606
10Qwen 3 VL 235B A22B Instruct28.6#532
11Qwen 3 VL 32B (Thinking)28.6#573
12GPT-5 MiniHigh24.5#364
13GLM-4.6V20.4#710
14Qwen 3 VL 8B Instruct16.3#1029
15O3High16.3#237

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

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