Paxa Labs and Typhoon OCR for Thai documents
Read Thai documents into Markdown and fill typed fields from them. How the two compare on languages, output, schema extraction, pages per request, limits, and self-hosting.
Both read Thai documents into Markdown. Typhoon OCR 1.5 is an open 2B model for English and Thai. Its API is a free research showcase capped at 20 requests a minute, and its weights run on your own hardware. Paxa OCR Lite is a paid API that reads up to 50 pages a request, and Paxa Document Extraction fills a typed schema from the same documents.
Sources:Typhoon OCR 1.5 model card(opens in a new tab)Typhoon FAQ(opens in a new tab)Typhoon rate limits(opens in a new tab)
Rates and claims last checked October 2026
What differs
| Dimension | Paxa Labs | Typhoon OCR(opens in a new tab) |
|---|---|---|
| Languages | Paxa OCR Lite and Paxa Document Extraction read Thai and other languages. The request carries no language field. | English and Thai documents. |
| Reading output | Each page as GitHub-flavored Markdown in reading order, with headings, lists, and tables, or as typed blocks of the same reading. | Markdown for text, HTML for tables with merged cells, <figure> tags for figures and charts, LaTeX for equations, and <page_number> tags for page numbers. |
| Field extraction | Paxa Document Extraction fills the schema you send with each request, up to 100 leaf fields. Built-in types cover Thai identifiers checked by check digit, provinces, banks, and dates in either era. | Their OCR guide documents one output, structured and layout-aware Markdown. |
| Files and pages per request | PDF, PNG, JPEG, or WebP, detected from the bytes. Up to 50 pages a request for OCR and 20 for extraction. | PDF, PNG, and JPEG. Each call reads the one page named by page_num, and their guide leaves the loop over a longer document to your code. |
| Access and limits | A paid API priced per page. Requests a minute follow the plan, from 10 without one to no limit on the largest. | Their FAQ calls the API a free research showcase, provided as-is and in beta. OCR is capped at 2 requests a second and 20 a minute, and higher limits are by request. |
| Running it yourself | Served from our hosted API and console. | Open weights of 2B parameters under Apache-2.0 on Hugging Face. The model card recommends vLLM, or Ollama on device. |
Sources:Typhoon OCR 1.5 model card(opens in a new tab)Typhoon OCR guide(opens in a new tab)Typhoon FAQ(opens in a new tab)Typhoon rate limits(opens in a new tab)
- Paxa OCR Lite, per 1,000 pages
- $6.50
- Paxa Document Extraction, per 1,000 pages
- $13
- Typhoon OCR
- No published rate
Typhoon publishes no per-page rate. Its FAQ calls the API free to use as a research showcase in beta. Every rate is per 1,000 pages at the vendor's first pay-as-you-go volume band. Field extraction is compared with ours for a schema of up to 50 fields. A larger schema costs $20 per 1,000 pages.
When Typhoon OCR is the better choice
Typhoon OCR costs nothing on their API within 20 requests a minute. Its open weights under Apache-2.0 run on your own GPUs with vLLM, or on device with Ollama. Its model card lists handwriting, receipts, and government forms, and it returns tables as HTML with merged cells and equations as LaTeX. Choose us for a paid API that reads up to 50 pages a request and fills a typed schema from Thai documents, with ID numbers checked by their check digit.
Questions
Is Typhoon OCR free?
Their FAQ calls the Typhoon API a research showcase that is free to use, provided as-is and in beta. OCR on it is capped at 20 requests a minute, and the weights are open under Apache-2.0.
Can I run Typhoon OCR on my own servers?
Yes. The weights are on Hugging Face under Apache-2.0, and the model card recommends vLLM, or Ollama for use on device.
Does Typhoon OCR fill a schema?
Their OCR guide documents Markdown output. Paxa Document Extraction fills a schema you send, and each value is printed text or null with the reason.
Do you publish accuracy comparisons?
No. We publish no accuracy, quality, or speed figures of our own, for our models or anyone else's. Run the same documents through both and compare the output.
Other comparisons
- Best OCR APIs for Each Job
- Paxa Labs and Amazon Textract for OCR and document extraction$1.50-4.00
- Paxa Labs and Azure Document Intelligence for OCR and document extraction$1.50-10
- Paxa Labs and Mistral OCR for reading and extracting documents$2-4
- Paxa Labs and LlamaParse for OCR and document extraction$1.25-56
Start with free credits
Sign in with Google, GitHub, or Hugging Face and spend 100 one-time free credits on your own text.
Start free