WildTableBench - Semantic Lookup and Structure: leaderboard
Metric: Accuracy (%) on the Semantic Lookup and Structure subtype (C1-S), 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 | Kimi K2.5 (Thinking) | 56.7 |
| 2 | Seed 2.0 Pro (High) | 56.7 |
| 3 | Gemini 3 Flash | 50 |
| 4 | Gemini 3 Pro | 46.7 |
| 5 | GPT-5.2 (High) | 46.7 |
| 6 | Qwen 3 VL 32B (Thinking) | 46.7 |
| 7 | Claude Sonnet 4.6 (High) | 46.7 |
| 8 | Qwen 3 VL 235B A22B (Thinking) | 33.3 |
| 9 | Claude Opus 4.6 (High) | 33.3 |
| 10 | Qwen 3 VL 32B Instruct | 23.3 |
| 11 | Qwen 3 VL 235B A22B Instruct | 23.3 |
| 12 | GPT-5 Mini (High) | 20 |
| 13 | O3 (High) | 20 |
| 14 | GLM-4.6V | 10 |
| 15 | Qwen 3 VL 8B Instruct | 3.3 |
Interactive version: theaggregate.ai/benchmark?slug=wildtablebench-semantic-lookup-and-structure · How It Works · Data refreshed daily, snapshot 2026-10-07.