Chandra vs Phi-4-multimodal
Pick Chandra for scanned, multilingual or table-heavy documents where you need Markdown, HTML or JSON with bounding boxes. Pick Phi-4-multimodal for English document Q&A on a modest GPU, or when one model must handle audio and images. The main split is language and layout: Chandra covers 43+ languages and returns coordinates, while Phi-4-multimodal supports vision in English only and has no layout output.
Chandradatalab/chandra | Phi-4-multimodalmicrosoft/phi-4-multimodal | |
|---|---|---|
| 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 | 2.5 s / page |
| Best for | Complex layouts and handwriting | Small model for OCR and questions |
| Inputs | PDF, images | images |
| Languages | 40+ languages | English for vision |
| 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 | No |
| 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
- You need bounding boxes and labelled layout blocks, plus Markdown, HTML or JSON output.
- Your documents are not only English: Chandra 2 averages 77.8% across 43 languages and 72.7% on a 90-language benchmark.
- You want a hosted option: the Datalab API costs $1, $3 or $7.50 per 1,000 pages, with $20/month free.
Choose Phi-4-multimodal if
- Your documents are English and you mainly ask questions about them, where it scores 93.2 on DocVQA at 5.6B parameters.
- You need a permissive licence: MIT, with no revenue or funding cap for commercial use.
- You also process audio, or you need a 128K token context in the same model.
Key differences
| Chandra | Phi-4-multimodal | |
|---|---|---|
| Licence | Code Apache-2.0; weights modified OpenRAIL-M, free for research, personal use and companies under $2M funding/revenue | MIT |
| Cost | Self-hosted is free within licence terms; hosted API is $1, $3 or $7.50 per 1,000 pages | Free self-hosted (you pay for compute); also available via Azure AI Foundry |
| Hardware | Single NVIDIA GPU via vLLM; presets for A100, L40S and RTX 4090/5090 | Flash attention on A100, A6000 and H100; V100 and older need eager attention |
| Output | Markdown, HTML, JSON and bounding boxes | Plain text, Markdown and JSON; no coordinates or layout |
| Languages | 77.8% average across 43 languages; weak on Urdu (43.2%) and Sanskrit (51.1%) | Vision supported for English only; performance drops in other languages |
| Benchmarks | 85.8 on olmOCR-bench (Chandra 2) | 93.2 DocVQA, 84.4 OCRBench, 81.4 ChartQA, 72.7 InfoVQA |
| Known failure | Token-loop hallucinations on some Russian/Kazakh scans; weak on old degraded scans (51.1 on Old Scans) | Repetitive or inconsistent output in long chats |
Sources
What each one is
Chandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra detailsPhi-4-multimodal
Microsoft's 5.6B text, vision and audio model (February 2025). Scores 84.4 on OCRBench per its model card.
Phi-4-multimodal 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": ["microsoft/phi-4-multimodal"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Chandra alternativesPhi-4-multimodal alternativesBest OCR models in 2026