PorTEXTO - Synthetic: leaderboard
PorTEXTO is a European Portuguese visual-text-extraction benchmark over four subsets (handwritten region and full page, synthetic documents, in-the-wild images). This board reports ANLS (%) on the synthetic pt-PT document subset; higher is better.
Metric: ANLS (%) on the synthetic pt-PT document subset; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3.6-27B [28] | 99.9 | |
| 2 | Qwen3.5-9B [27] | 99.8 | |
| 3 | Ministral3-Instruct-8B [13] | 98.6 | |
| 4 | Gemma-4-31B-it [8] | 98.3 | |
| 5 | LLaVA-OneVision-1.5-8B [1] | 98.2 | |
| 6 | Gemma-4-E4B-it [8] | 95.6 | |
| 7 | Bee-RL-8B [37] | 94.6 | |
| 8 | Nemotron-Nano-12B-VL-V2 [5] | 92.2 | |
| 9 | Nemotron-OCR-V2 [14] | 91.4 | |
| 10 | DeepSeek-OCR-2-3B [33] | 91.4 |
Interactive version: theaggregate.ai/benchmark?slug=portexto-synthetic · How It Works · Data refreshed daily, snapshot 2026-11-13.