WildTableBench - Multi-step Conditional: leaderboard

A WildTableBench (2026) subtype on a multi-step numeric result under several nested conditions, 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 Multi-step Conditional subtype (C2-M), 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 Pro66.4#146
2Gemini 3 Flash51#180
3GPT-5.2High49#190
4Kimi K2.5Thinking45.5#272
5Claude Opus 4.6High45.5#114
6Seed 2.0 ProHigh42#226
7Claude Sonnet 4.6High34.3#179
8Qwen 3 VL 235B A22B (Thinking)32.2#462
9Qwen 3 VL 235B A22B Instruct25.2#541
10Qwen 3 VL 32B Instruct23.8#607
11GLM-4.6V23.8#721
12Qwen 3 VL 32B (Thinking)23.8#577
13GPT-5 MiniHigh21.7#373
14O3High14#242
15Qwen 3 VL 8B (Thinking)8.4#924

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

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