WildTableBench - Row Operation: leaderboard

A WildTableBench (2026) subtype on recomputing a result after rows are added or removed under a condition, 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 Row Operation subtype (C4-R), 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 Pro77.5#145
2Claude Opus 4.6High67.5#113
3Gemini 3 Flash57.5#175
4Kimi K2.5Thinking57.5#266
5Claude Sonnet 4.6High50#173
6Seed 2.0 ProHigh47.5#222
7GPT-5.2High45#185
8GPT-5 MiniHigh37.5#364
9GLM-4.6V35#710
10Qwen 3 VL 32B (Thinking)35#573
11Qwen 3 VL 235B A22B (Thinking)30#459
12Qwen 3 VL 32B Instruct27.5#606
13Qwen 3 VL 235B A22B Instruct25#532
14Qwen 3 VL 8B (Thinking)20#910
15Qwen 3 VL 8B Instruct15#1029

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

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