SmolDocling vs Phi-4-multimodal
Pick Phi-4-multimodal for English document Q&A and extraction on a single modern GPU, or if you need an MIT licence. Pick SmolDocling only for cheap experiments that need page conversion with bounding boxes. SmolDocling is a preview that its authors say is replaced by Granite-Docling, so it is a poor fit for production.
SmolDoclinghuggingface/smoldocling | Phi-4-multimodalmicrosoft/phi-4-multimodal | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.0 s / page | 2.5 s / page |
| Best for | Very cheap page conversion | Small model for OCR and questions |
| Inputs | PDF, images | images |
| Languages | Mainly English | English for vision |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | 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 SmolDocling if
- You need layout output: DocTags, bounding boxes, HTML, JSON, Markdown and LaTeX from one 256M-parameter model.
- You want to run page conversion on modest hardware, at about 0.35 seconds per page on an A100 with vLLM.
- You are experimenting with tiny document VLMs and can accept preview-quality output.
Choose Phi-4-multimodal if
- You want to ask questions about English documents, where it scores 93.2 on DocVQA.
- You need a permissive MIT licence for commercial use, with a hosted option on Azure AI Foundry.
- Your pipeline mixes audio and images, or needs a 128K token context in one model.
Key differences
| SmolDocling | Phi-4-multimodal | |
|---|---|---|
| Size | 256M parameters | 5.6B parameters |
| Licence | Apache-2.0 per the model summary (the metadata tag says CDLA-Permissive-2.0) | MIT |
| How you run it | Self-host only | Self-host, or use Azure AI Foundry |
| Output formats | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes | Plain text, Markdown, JSON. No coordinates or layout. |
| OCRBench | 338 | 84.4 |
| Languages | English only | Vision supported for English only; performance drops in other languages |
| Known failure | Endless token repetition and missing closing tags; output can loop on some images and make generation very slow | Repetitive or inconsistent output in long chats; V100 and older GPUs need the eager attention fallback |
Sources
- SmolDocling model card
- SmolDocling paper
- Docling issue #1515
- Granite-Docling comparison table
- Hugging Face model card (Phi-4-multimodal)
What each one is
SmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling 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": "huggingface/smoldocling","fallbacks": ["microsoft/phi-4-multimodal"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
SmolDocling alternativesPhi-4-multimodal alternativesBest OCR models in 2026