Jina OCR vs InternVL3.5
Pick Jina OCR for non-commercial page-to-Markdown work where you want a published OmniDocBench score and a hosted API option. Pick InternVL3.5 for commercial or self-hosted pipelines that need an Apache-2.0 licence and a choice of model sizes. The main split is licence and scope: Jina OCR is a dedicated OCR model under CC-BY-NC-4.0, while InternVL3.5 is a general vision language model with no text OCR scores published.
Jina OCRjina/jina-ocr | InternVL3.5opengvlab/internvl | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.5 s / page | 3.0 s / page |
| Best for | Clean Markdown at low cost | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | About 100 languages | Multilingual |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ 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 Jina OCR if
- It scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, 7.4 points above DeepSeek-OCR.
- It activates only about 570M parameters of a 3B MoE and runs at 2.57 pages per second on an A100.
- You can call Jina's hosted API instead of running your own GPUs, and it outputs Markdown, HTML and LaTeX.
Choose InternVL3.5 if
- The Apache-2.0 licence allows commercial use without a separate deal.
- Nine sizes from 1B to 241B let you trade accuracy for cost, and models up to 30B fit on a single A100.
- It suits document question answering and multimodal pipelines, since it is a general vision language model that can answer questions about a page.
Key differences
| Jina OCR | InternVL3.5 | |
|---|---|---|
| Licence | CC-BY-NC-4.0, commercial use needs a licence from Jina/Elastic | Apache-2.0 |
| How you run it | Self-host or Jina's hosted API | Self-host only (Transformers, LMDeploy or vLLM) |
| Model size | 3B MoE, about 570M active | 1B to 241B-A28B, nine sizes |
| Hardware | GPU recommended (A100, L4); falls back to CPU in Transformers | Up to 30B on one A100; 38B needs two A100s; 241B-A28B needs eight A100s |
| Output formats | Markdown, HTML, LaTeX | Plain text, Markdown, JSON |
| OCR benchmarks | OmniDocBench v1.6 91.14, olmOCR-Bench 83.4 (authors' figures) | Not published as text values; the model card shows charts only |
| Known failure | Hosted API returns HTTP 503 on cold start; retry after 30 to 60 seconds | No bounding boxes or layout output by default |
Sources
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina 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": "jina/jina-ocr","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesInternVL3.5 alternativesBest OCR models in 2026