OCRTurk - Turkish Characters: leaderboard
Metric: Turkish character sensitivity (0-1): one minus the share of the Turkish-specific letters (c-cedilla, g-breve, dotless i, o-umlaut, s-cedilla, u-umlaut and their capitals) in the reference text that the output gets wrong, averaged over the 180 pages; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | HunyuanOCR | 0.88 | |
| 2 | OCRTurk PaddleOCR-VL (checkpoint unspecified) | 0.82 | |
| 3 | DeepSeek-OCR | 0.81 | |
| 4 | olmOCR-2 | 0.8 | |
| 5 | Nanonets-OCR2-3B | 0.79 | |
| 6 | Docling | 0.71 | |
| 7 | NVIDIA-Nemotron-Parse-v1.1 | 0.47 |
Interactive version: theaggregate.ai/benchmark?slug=ocrturk-turkish-characters · How It Works · Data refreshed daily, snapshot 2026-10-11.