Typhoon OCR vs InternVL3.5
Pick Typhoon OCR for Thai and Thai-English documents such as government forms, financial reports and books, where a 2B model scores 84.7 on olmOCR-bench. Pick InternVL3.5 if you want a general vision language model in a size that fits your hardware, mainly for document question answering. Typhoon OCR is built for OCR output (Markdown, HTML tables, LaTeX), while InternVL3.5 is not.
Typhoon OCRtyphoon/typhoon-ocr | InternVL3.5opengvlab/internvl | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 1.8 s / page | 3.0 s / page |
| Best for | Thai documents | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | Thai, English | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose Typhoon OCR if
- Your documents are Thai or Thai-English, such as government forms, financial reports or books. V1.5 beat GPT-5 and Gemini 2.5 Pro on Thai government forms (BLEU 0.870).
- You need tables as HTML with merged cells, equations as LaTeX and checkboxes marked.
- You want published OCR benchmark numbers at a small size: 84.7 on olmOCR-bench and 90.97 on OmniDocBench v1.6 with the 2B V2 model.
Choose InternVL3.5 if
- You want one model family in nine sizes from 1B to 241B, so you can trade accuracy for cost.
- Your job is document question answering or a wider self-hosted multimodal pipeline, not only page-to-text conversion.
- You need a clear hardware plan: models up to 30B run on a single A100, 38B needs two A100s and 241B-A28B needs eight.
Key differences
| Typhoon OCR | InternVL3.5 | |
|---|---|---|
| Licence | Apache-2.0, subject to OpenTyphoon Terms and Conditions | Apache-2.0 |
| How you run it | Self-host the weights or use the Typhoon API (typhoon-ocr-v2) | Self-host only, with Transformers, LMDeploy or vLLM |
| Model size | 2B (V2 and V1.5), plus a 0.8B V2 variant | 1B to 241B-A28B; the 8B variant is 8.5B |
| Output formats | Markdown, HTML, LaTeX | Plain text, Markdown, JSON; no bounding boxes or layout output by default |
| Published OCR scores | ThaiOCRBench 59.9, olmOCR-bench 84.7, OmniDocBench v1.6 90.97 (V2 2B) | Not published as text; the model card shows OCR scores only as chart images |
| Languages evaluated | Thai and English only | Not published |
| Known failure | V2 can loop until the token limit, mostly on dense multi-column pages. Key-information extraction is weakest (21.0 on ThaiOCRBench). | Not suited to jobs needing word coordinates or CPU-only runs of larger sizes |
Sources
- Typhoon OCR V2 2B model card
- Typhoon OCR technical report
- Typhoon OCR 1.5 release blog
- Hugging Face model card (8B)
- InternVL GitHub
What each one is
Typhoon OCR
Open vision model tuned for Thai and English documents, with Markdown output.
Typhoon OCR detailsInternVL3.5
Shanghai AI Lab's open vision-language family (1B to 241B, August 2025), strong on OCR and document benchmarks.
InternVL3.5 detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "typhoon/typhoon-ocr","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Typhoon OCR alternativesInternVL3.5 alternativesBest OCR models in 2026