WildTableBench - Hypothetical: leaderboard

Accuracy in percent on the hypothetical category: results recomputed after removing rows, editing values or assuming a what-if condition, free-form questions about real-world table images collected from the web (WildTableBench, 2026), answers judged against the reference by GPT-5.2. Higher is better.

Metric: Accuracy (%) on the Hypothetical question category (C4, question-weighted over its subtypes), 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 Pro75.3#146
2Gemini 3 Flash55.1#180
3Kimi K2.5Thinking55.1#272
4GPT-5.2High55.1#190
5Claude Opus 4.6High53.9#114
6Claude Sonnet 4.6High49.4#179
7Seed 2.0 ProHigh44.9#226
8Qwen 3 VL 235B A22B (Thinking)39.3#462
9GPT-5 MiniHigh33.7#373
10Qwen 3 VL 32B Instruct27#607
11GLM-4.6V27#721
12Qwen 3 VL 235B A22B Instruct21.3#541
13Qwen 3 VL 32B (Thinking)20.2#577
14Qwen 3 VL 8B (Thinking)16.9#924
15O3High10.1#242

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

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