FormStruct-Bench - Value Similarity: leaderboard
Metric: Value-nED (%; normalized Levenshtein similarity of non-empty leaf values under maximum-weight one-to-one matching, divided by the larger value count; 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 March 2027. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 80.98 |
| 2 | Seed 2.1 Pro | 77.45 |
| 3 | Kimi K2.6 | 76.63 |
| 4 | Claude Sonnet 5 | 75.42 |
| 5 | Qwen 3.7 Plus | 74.02 |
| 6 | Qwen 3.6 35B A3B | 66.93 |
| 7 | Gemini 3.5 Flash | 66.8 |
| 8 | Qwen 3.5 9B | 65.73 |
| 9 | Step3 VL 10B | 59.48 |
Interactive version: theaggregate.ai/benchmark?slug=formstruct-bench-value-similarity · How It Works · Data refreshed daily, snapshot 2026-09-26.