WildTableBench - Numerical: leaderboard

Accuracy in percent on the numerical category: basic arithmetic, ranking, conditional aggregates and multi-step conditional computation, 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 Numerical question category (C2, 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 Pro71.9#146
2Kimi K2.5Thinking50.6#272
3Gemini 3 Flash49.7#180
4Claude Opus 4.6High46.7#114
5GPT-5.2High45.8#190
6Seed 2.0 ProHigh44.9#226
7Claude Sonnet 4.6High34.1#179
8Qwen 3 VL 235B A22B (Thinking)33.5#462
9Qwen 3 VL 32B (Thinking)27.8#577
10Qwen 3 VL 235B A22B Instruct26.6#541
11GPT-5 MiniHigh24.3#373
12GLM-4.6V24#721
13Qwen 3 VL 32B Instruct22.5#607
14Qwen 3 VL 8B (Thinking)15#924
15O3High11.4#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-numerical · How It Works · Data refreshed daily, snapshot 2026-11-22.