OCRTurk - Equations: leaderboard
Metric: Character detection matching (0-1), the F1 of rendered formula characters matched between the output and reference LaTeX, over the benchmark's 92 equations; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | OCRTurk PaddleOCR-VL (checkpoint unspecified) | 0.95 | |
| 2 | Nanonets-OCR2-3B | 0.95 | |
| 3 | DeepSeek-OCR | 0.92 | |
| 4 | HunyuanOCR | 0.89 | |
| 5 | olmOCR-2 | 0.88 | |
| 6 | NVIDIA-Nemotron-Parse-v1.1 | 0.86 | |
| 7 | Docling | 0.85 |
Interactive version: theaggregate.ai/benchmark?slug=ocrturk-equations · How It Works · Data refreshed daily, snapshot 2026-10-11.