WildTableBench - Cell-Level: leaderboard
Metric: Accuracy (%) on the Cell-Level question category (C1, 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 January 2027. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 62.8 |
| 2 | Seed 2.0 Pro (High) | 59.3 |
| 3 | GPT-5.2 (High) | 53.1 |
| 4 | Claude Opus 4.6 (High) | 51.7 |
| 5 | Gemini 3 Flash | 51 |
| 6 | Kimi K2.5 (Thinking) | 47.6 |
| 7 | Qwen 3 VL 235B A22B (Thinking) | 37.9 |
| 8 | Claude Sonnet 4.6 (High) | 35.9 |
| 9 | Qwen 3 VL 32B (Thinking) | 28.3 |
| 10 | Qwen 3 VL 32B Instruct | 24.8 |
| 11 | Qwen 3 VL 235B A22B Instruct | 22.8 |
| 12 | GPT-5 Mini (High) | 20.7 |
| 13 | GLM-4.6V | 17.9 |
| 14 | O3 (High) | 17.2 |
| 15 | Qwen 3 VL 8B Instruct | 10.3 |
Interactive version: theaggregate.ai/benchmark?slug=wildtablebench-cell-level · How It Works · Data refreshed daily, snapshot 2026-10-07.