WildTableBench - Transcription: leaderboard

A WildTableBench (2026) subtype on listing the cell values of a given row or column range in order, 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 Transcription subtype (C1-T), 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 Pro66.7#146
2Seed 2.0 ProHigh50#226
3Kimi K2.5Thinking46.7#272
4Claude Opus 4.6High46.7#114
5Gemini 3 Flash43.3#180
6GPT-5.2High43.3#190
7GLM-4.6V26.7#721
8Qwen 3 VL 235B A22B (Thinking)26.7#462
9Qwen 3 VL 235B A22B Instruct23.3#541
10GPT-5 MiniHigh23.3#373
11Claude Sonnet 4.6High20#179
12Qwen 3 VL 32B Instruct16.7#607
13Qwen 3 VL 32B (Thinking)13.3#577
14Qwen 3 VL 8B Instruct10#1043
15Qwen 3 VL 4B Instruct10#1348

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

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