WildTableBench - Aggregate Verification: leaderboard

A WildTableBench (2026) subtype on judging a claimed sum, average or count 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 Aggregate Verification subtype (C3-A), 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 K2.5Thinking81.8#272
2Seed 2.0 ProHigh63.6#226
3Gemini 3 Flash54.5#180
4Gemini 3 Pro54.5#146
5GPT-5.2High45.5#190
6Claude Opus 4.6High45.5#114
7Qwen 3 VL 235B A22B Instruct36.4#541
8Qwen 3 VL 32B Instruct36.4#607
9Qwen 3 VL 235B A22B (Thinking)36.4#462
10Claude Sonnet 4.6High36.4#179
11GLM-4.6V27.3#721
12GPT-5 MiniHigh27.3#373
13Qwen 3 VL 8B Instruct18.2#1043
14Qwen 3 VL 8B (Thinking)18.2#924
15Qwen 3 VL 32B (Thinking)18.2#577

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

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