Chandra vs Qwen-VL
Pick Chandra for scanned documents, multilingual pages, and PDF to Markdown for RAG, where faithful transcription matters. Pick Qwen-VL when you want one Apache-2.0 model that also answers questions about images, photos and screenshots. The main split is accuracy: Chandra 2 scores 85.8 on olmOCR-bench, while Qwen 3 VL 8B scores 64.6.
Chandradatalab/chandra | Qwen-VLqwen/qwen-vl | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 2.0 s / page | 2.0 s / page |
| Best for | Complex layouts and handwriting | Flexible extraction with prompts |
| Inputs | PDF, images | PDF, images |
| Languages | 40+ languages | 30+ languages |
| Max pages per request | 500 | 200 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | ✓ Yes |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| 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 scores 85.8 on olmOCR-bench (Chandra 2), well above the 64.6 reported for Qwen 3 VL 8B.
- It averages 77.8% across 43 languages and returns Markdown, HTML or JSON with bounding boxes per block.
- You can self-host it or use the Datalab API at $1, $3 or $7.50 per 1,000 pages, with $20 a month free.
Choose Qwen-VL if
- Its Apache-2.0 licence has no revenue limit, while Chandra's weights need a paid licence for companies above $2M funding or revenue.
- It comes in sizes from 2B to 235B-A22B MoE, so you can match the model to your GPU budget.
- It is a general model, so one deployment can do OCR, reasoning and visual question answering on photos and screenshots.
Key differences
| Chandra | Qwen-VL | |
|---|---|---|
| Licence | Code Apache-2.0; weights modified OpenRAIL-M, free under $2M funding/revenue, paid above that | Apache-2.0 |
| How you run it | Self-host with vLLM or Hugging Face, or the hosted Datalab API | Self-host open weights, or hosted on Alibaba Cloud; hosted pricing not verified |
| Price | Free self-hosted within licence terms; API $1, $3 or $7.50 per 1,000 pages | Free self-hosted, you pay for compute; hosted price varies by model |
| olmOCR-bench | 85.8 overall (Chandra 2) | 64.6 (Qwen 3 VL 8B) |
| Languages | 43-language average of 77.8%; weak on Urdu (43.2%) and Sanskrit (51.1%) | OCR in 32 languages |
| Output formats | Markdown, HTML, JSON, bounding boxes | Plain text, HTML, bounding boxes, JSON, Markdown |
| Known failure | Token loops reported on 5-6 of about 60 Russian/Kazakh scans; weak on old degraded scans | Repeating-text loops, worse with greedy decoding and on low-quality images; loops on 10-37% of outputs with transformers 5.17 |
Sources
- Chandra GitHub README
- Datalab pricing
- GitHub issue #51 token loops
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- olmo-eval issue #441
What each one is
Chandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra detailsQwen-VL
Alibaba's open-weight vision language models (Qwen2.5-VL, Qwen3-VL). Strong general OCR and grounding.
Qwen-VL detailsTry 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": "datalab/chandra","fallbacks": ["qwen/qwen-vl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Chandra vs Granite DoclingChandra vs Hunyuan OCRSurya vs Qwen-VLolmOCR vs Qwen-VLQwen-VL vs MonkeyOCRMathpix Convert vs Qwen-VLQwen-VL vs Phi-4-multimodalQwen-VL vs OpenOCR
Chandra alternativesQwen-VL alternativesBest OCR models in 2026