Skip to main content

We did it! Thai document AI, built in Thailand, two models free: OpenThai 2.0 OCR

We did it! Thai document AI, two free models

Update, 18 Nov 2026: OpenThai 2.1 OCR is out. It reads phone photos of documents much better and fixes what users reported. See what's new. The Hugging Face links in this post still work and lead to 2.1.

We did it! Thai document AI, built by a Thai team, with two models free to download: OpenThai 2.0 OCR Markdown and OpenThai 2.0 OCR Text. Commercial use is allowed.

Sound familiar? You scan a stack of government documents, the OCR reads every table as one long line, you can no longer tell which number belongs to which row, and you end up retyping it cell by cell.

1) OpenThai 2.0 OCR Markdown reads a page into Markdown, and tables stay tables.

  • 95.2 on 65 real Thai government and university pages with tables (Typhoon OCR V2 2B: 91.6).
  • 98.7% character accuracy on 339 Royal Gazette pages published after training finished.
  • 90.3 average over 12 printed Thai text sets, the highest of the 15 engines we tested.
  • 1.6 s per page on one NVIDIA H100. 0.63B, fine-tuned from NVIDIA Nemotron 3.5 Parse.

2) OpenThai 2.0 OCR Text is a 4 MB model that needs no GPU.

  • 0.39 s per page on an ordinary CPU, about 950 pages a minute on 8 cores.
  • Works offline, so confidential documents never leave the machine.
  • 99.5 on budget documents and 96.3 on the Royal Gazette (Gemini 2.5 Flash: 81.0 and 92.6; Google Cloud Vision: 75.0 and 87.8).

Why free? Millions of Thai pages in government offices, hospitals and companies cannot be sent to a foreign cloud. We want every organisation to be able to read them on its own machines. And anyone building an AI chatbot on their own documents has to start by reading those documents correctly.

Honestly: fully handwritten pages and photos of short words are not our strength yet; on single-word photos Typhoon OCR still does better. The team is working on it. Thanks to the Typhoon team for raising the bar for Thai OCR with us.

Download: huggingface.co/iapp/OpenThai2.0-OCR-Markdown and huggingface.co/iapp/OpenThai2.0-OCR-Text API: OCR Markdown API (0.15 IC per page) and OCR Text API (plain text 0.049 IC per page) Full results, usage and limits: OpenThai 2.0 OCR model page

Built on NVIDIA Nemotron · trained on NVIDIA H100 GPUs provided by Siam AI