UniKIE-Bench (Constrained-Category): leaderboard
Metric: Field-level F1 (%) averaged over the 11 application scenarios of UniKIE-Bench's constrained-category track (4,472 real documents from public KIE datasets, each scenario with predefined extraction schemas, in business, public-service and regulated-record domains); schema-guided key information extraction from a document image in one pass; field-level F1, a field correct only when its value matches the annotation exactly; one prompt for all models, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 82.37 | #64 |
| 2 | Qwen 3 VL 8B Instruct | 79.12 | #401 |
| 3 | GLM-4.1V-9B (Thinking) | 69.91 | #457 (GLM-4.1V-9B) |
| 4 | GPT-4o | 69.15 | #333 |
| 5 | GPT-5 | 67.59 | #91 |
| 6 | Claude Sonnet 4.5 | 66.39 | #138 |
| 7 | Gemma 3 12B (IT) | 48.09 | #655 |
| 8 | Ministral-3-8B-Instruct-2512 | 37.62 | #644 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=unikie-bench-constrained-category · How It Works · Data refreshed daily, snapshot 2026-10-11.