WildTableBench - Aggregate Verification: leaderboard
Metric: Accuracy (%) on the Aggregate Verification subtype (C3-A), 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) | 81.8 |
| 2 | Seed 2.0 Pro (High) | 63.6 |
| 3 | Gemini 3 Flash | 54.5 |
| 4 | Gemini 3 Pro | 54.5 |
| 5 | GPT-5.2 (High) | 45.5 |
| 6 | Claude Opus 4.6 (High) | 45.5 |
| 7 | Qwen 3 VL 32B Instruct | 36.4 |
| 8 | Qwen 3 VL 235B A22B Instruct | 36.4 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 36.4 |
| 10 | Claude Sonnet 4.6 (High) | 36.4 |
| 11 | GLM-4.6V | 27.3 |
| 12 | GPT-5 Mini (High) | 27.3 |
| 13 | Qwen 3 VL 8B Instruct | 18.2 |
| 14 | Qwen 3 VL 8B (Thinking) | 18.2 |
| 15 | Qwen 3 VL 32B (Thinking) | 18.2 |
Interactive version: theaggregate.ai/benchmark?slug=wildtablebench-aggregate-verification · How It Works · Data refreshed daily, snapshot 2026-10-07.