OCRRouter

InternVL3.5 vs Unstructured

Pick InternVL3.5 if you want to run a general vision language model on your own GPUs under Apache-2.0 and ask questions about document images. Pick Unstructured if you need to ingest many file types (PDF, HTML, Word, PowerPoint, email, images) into typed elements with bounding boxes for RAG. The main split is that InternVL3.5 is a model you host, while Unstructured is a pipeline with a hosted API.

InternVL3.5opengvlab/internvlUnstructuredunstructured/unstructured
TypeVision modelOCR API
Price2 credits / page20 credits / page
Per 1,000 pages$2.00$20.00
Typical speed3.0 s / page3.0 s / page
Best forSelf-hosted document understandingMixed file types into RAG
InputsimagesPDF, images, DOCX
LanguagesMultilingualMultilingual
Max pages per request5001,000
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ YesNo
Bounding boxesNo✓ Yes
ConfidenceNoNo
FormsNoNo
Multilingual✓ Yes✓ Yes
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose InternVL3.5 if

  • You want one model family in nine sizes, from 1B to 241B-A28B, to trade accuracy for cost.
  • You want document question answering and Markdown or JSON output from a self-hosted multimodal pipeline.
  • You want Apache-2.0 weights with no per-page fee, and models up to 30B fit on a single A100.

Choose Unstructured if

  • You need to ingest PDFs, HTML, Word, PowerPoint, emails and images through one library.
  • You need bounding boxes in the output, which InternVL3.5 does not give by default.
  • You want a hosted API with the first 10,000 pages free, then $0.015 per page.

Key differences

InternVL3.5Unstructured
How you run itSelf-host only, with Transformers, LMDeploy or vLLMSelf-host the Python library or call the hosted API
LicenceApache-2.0Apache-2.0 for the open-source library; proprietary for the hosted platform
PriceFree, you pay for computeLibrary is free. Hosted: first 10,000 pages free, then $0.015 per page. Business is custom.
Output formatsPlain text, Markdown, JSONJSON, HTML, plain text, bounding boxes
Layout and coordinatesNo bounding boxes or layout output by defaultReturns typed elements with bounding boxes
HardwareUp to 30B on one A100; 38B needs two A100s; 241B-A28B needs eight A100sNot published for the library; the hosted API needs no GPU on your side
Known failureOCR scores are published only as chart images, not text valuesOSS table cell content scores 0.426 versus 0.866 hosted; hi_res tables can lose first letters at cell edges, and table OCR ignores language settings

Sources

What each one is

InternVL3.5

Shanghai AI Lab's open vision-language family (1B to 241B, August 2025), strong on OCR and document benchmarks.

InternVL3.5 details

Unstructured

Document ETL for RAG. Partitions many file types into typed elements; open-source library too.

Unstructured details

Try both on your own file

Run the file once with each model and compare the output.

PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?

Use one, fall back to the other

{"model": "opengvlab/internvl","fallbacks": ["unstructured/unstructured"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

InternVL3.5 alternativesUnstructured alternativesBest OCR models in 2026