WildTableBench - Conditional Verification: leaderboard

A WildTableBench (2026) subtype on judging a conditional or filtered claim 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 Conditional Verification subtype (C3-C), 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 Pro75.8#145
2Gemini 3 Flash69.7#175
3Kimi K2.5Thinking69.7#266
4Seed 2.0 ProHigh66.7#222
5GPT-5.2High63.6#185
6Claude Opus 4.6High57.6#113
7Qwen 3 VL 235B A22B (Thinking)54.5#459
8Claude Sonnet 4.6High51.5#173
9GLM-4.6V48.5#710
10Qwen 3 VL 32B (Thinking)48.5#573
11Qwen 3 VL 32B Instruct42.4#606
12GPT-5 MiniHigh39.4#364
13Qwen 3 VL 235B A22B Instruct36.4#532
14GPT-4o33.3#751
15Qwen 3 VL 4B (Thinking)30.3#1334

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

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