Four Lightweight OCR Models You Can Run Locally on a Laptop
vanstriendaniel · x · 2026-08-09
The author shares four OCR models that can be run locally on a laptop, along with their respective links:
- surya-ocr-2: 0.7B parameters
- OvisOCR2: 0.9B parameters
- PaddleOCR-VL-1.6: 1B parameters
- dots.mocr: 3B parameters
All models are compatible with llama.cpp or MLX. The author specifically recommends OvisOCR2, noting that despite being somewhat under the radar, it delivers excellent results. For simple text extraction, the classic Tesseract engine also remains a viable option.
Related event: A Roundup of Four Lightweight OCR Models for Local Laptops(2 posts)→
More from Models
- Vision Test Fail: GPT Lags Behind Opus and Gemini in Low-Quality Image Recognition — stevejuniormc · 2026-08-09
- Dev Spent $20/Week on DeepSeek, Says It Actually Listens Unlike Codex — yacineMTB · 2026-08-09
- OpenAI's Next Major Model Codenamed 'Doug', Promising Massive Pre-Training Scale — Hesamation · 2026-08-09
- Google cancels imminent release plan; Gemma 4 to feature 120B parameters — legit_api · 2026-08-09
- OpenAI's Mysterious 'Doug' Codename: Unraveling the Astra Connection — koltregaskes · 2026-08-09
- GLint: A 150M Parameter SOTA Late-Interaction Retriever — IgorCarron · 2026-08-09