Chandra vs InternVL3.5
Pick Chandra when you need document OCR that returns Markdown, HTML or JSON with bounding boxes, especially for multilingual scans, tables, forms and math. Pick InternVL3.5 when you want an Apache-2.0 general vision model in a size you choose, for document question answering inside a self-hosted pipeline. The main split is purpose: Chandra is built for page conversion with layout, InternVL3.5 is a general VLM with no layout output by default.
Chandradatalab/chandra | 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 | 2.0 s / page | 3.0 s / page |
| Best for | Complex layouts and handwriting | Self-hosted document understanding |
| Inputs | PDF, images | images |
| Languages | 40+ languages | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | ✓ Yes | 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 Chandra if
- It outputs Markdown, HTML and JSON plus layout chunks with bounding boxes and labels per block.
- Its authors report 85.8 on olmOCR-bench (Chandra 2) and a 77.8% average across 43 languages.
- You can skip GPU hosting and use the Datalab API at $1, $3 or $7.50 per 1,000 pages, with a free tier of $20 per month.
Choose InternVL3.5 if
- Its Apache-2.0 licence has no revenue or funding limit, so large companies can self-host it commercially.
- It comes in nine sizes from 1B to 241B-A28B, so you can trade accuracy for cost.
- Models up to 30B fit on a single A100, and the 20B-A4B and 30B-A3B MoE variants activate few parameters per token.
Key differences
| Chandra | InternVL3.5 | |
|---|---|---|
| Licence | Code Apache-2.0; weights modified OpenRAIL-M. Free for research, personal use and companies under $2M funding or revenue; paid licence above that. | Apache-2.0 |
| How you run it | Self-host with vLLM or Hugging Face, or use the hosted Datalab API. | Self-host only, with Transformers, LMDeploy or vLLM. |
| Price | Self-hosted: free within licence terms. Hosted: $1, $3 or $7.50 per 1,000 pages; $20/month free tier. | Free, self-hosted. You pay for compute. |
| Output formats | Markdown, HTML, JSON and bounding boxes. | Plain text, Markdown and JSON. No bounding boxes or layout output by default. |
| OCR benchmark | 85.8 on olmOCR-bench (Chandra 2), reported by the authors. | Not published as text values; the model card shows OCR scores only as chart images. |
| Hardware | One NVIDIA GPU via vLLM. Benchmarks ran on one H100 80GB; presets for A100, L40S and RTX 4090/5090. | Up to 30B on one A100; 38B needs two A100s; 241B-A28B needs eight A100s. |
| Known failure | Token-loop hallucinations reported on 5-6 of about 60 Russian/Kazakh scans. Weak on Urdu (43.2%) and Old Scans (51.1 on olmOCR-bench). | No OCR failure data published. Not suited to jobs needing word coordinates. |
Sources
- Chandra GitHub README
- Datalab pricing
- GitHub issue #51 token loops
- Hugging Face model card (8B)
- InternVL GitHub
What each one is
Chandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra 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": "datalab/chandra","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Chandra alternativesInternVL3.5 alternativesBest OCR models in 2026