FormStruct-Bench - Path Accuracy: leaderboard
Metric: TSR-path (%; share of ground-truth leaf field-path and value pairs reproduced exactly, a strict leaf recall; macro-averaged over the 1,100 human-reviewed pages of the template-disjoint test split of table-form documents in seven languages; the model outputs the page as FormStruct-Bench JSON (deterministic decoding where available; pipelines as released), and missing, unparsable or schema-invalid outputs score 0). Source: arxiv.org. Saturation forecast: Around 2030. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.6 35B A3B | 17.91 |
| 2 | GPT-5.5 | 17.15 |
| 3 | Seed 2.1 Pro | 15.9 |
| 4 | Gemini 3.5 Flash | 14.55 |
| 5 | Claude Sonnet 5 | 13.77 |
| 6 | Kimi K2.6 | 12.96 |
| 7 | Qwen 3.5 9B | 12.58 |
| 8 | Qwen 3.7 Plus | 9.75 |
| 9 | Step3 VL 10B | 0.04 |
Interactive version: theaggregate.ai/benchmark?slug=formstruct-bench-path-accuracy · How It Works · Data refreshed daily, snapshot 2026-09-26.