Qwen-VL vs SmolDocling
Pick Qwen-VL when OCR is one part of a larger job, such as photos, screenshots, reasoning or VQA, or when you need OCR in many languages. Pick SmolDocling only for English page-to-Docling conversion on a tiny budget or for experiments, since it is a 256M preview that its authors say is superseded by Granite-Docling. Both can loop on some inputs, so check outputs either way.
Qwen-VLqwen/qwen-vl | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 2.0 s / page | 1.0 s / page |
| Best for | Flexible extraction with prompts | Very cheap page conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 30+ languages | Mainly English |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | 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 Qwen-VL if
- You need OCR beyond English: Qwen3-VL reads 32 languages, while SmolDocling is English only.
- You want one general model for OCR plus reasoning or VQA, on photos and screenshots with text, with sizes from 2B to 235B-A22B under Apache-2.0.
- You can use a hosted API: the open weights are also served on Alibaba Cloud, so you do not have to run GPUs yourself.
Choose SmolDocling if
- You have modest hardware: at 256M parameters it runs at about 0.35 seconds per page on an A100 with vLLM.
- You want structured page conversion in one pass: DocTags output with tables (OTSL), code, formulas and location tags, convertible to Markdown, HTML, JSON and LaTeX.
- You are experimenting with tiny document models on English pages and do not need production support.
Key differences
| Qwen-VL | SmolDocling | |
|---|---|---|
| Licence | Apache-2.0 (Qwen3-VL repository) | Apache-2.0 per the model summary; the metadata tag says CDLA-Permissive-2.0 |
| How you run it | Self-host the open weights or use the hosted API on Alibaba Cloud | Self-host only |
| Model size | 2B, 4B, 8B, 32B dense; 30B-A3B and 235B-A22B MoE | 256M |
| Languages | OCR in 32 languages | English only |
| Output formats | Plain text, HTML, bounding boxes, JSON, Markdown | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Published benchmarks | 64.6 on olmOCR-bench for Qwen 3 VL 8B, in an independent table, below dedicated OCR models | Table TEDS 0.82 on FinTabNet, full-page OCR F1 0.80, OCRBench 338, all reported by the authors; lower than Granite-Docling on every task |
| Known failure | Can loop repeating the same text, worse with greedy decoding and on low-quality scans; Qwen2.5-VL-32B GGUF often loops on scanned documents | Endless token repetition and missing closing tags; Docling issue #1515 reports looping that makes generation very slow even on sparse pages |
Sources
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- Chandra README benchmark table
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
What each one is
Qwen-VL
Alibaba's open-weight vision language models (Qwen2.5-VL, Qwen3-VL). Strong general OCR and grounding.
Qwen-VL detailsSmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling 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": "qwen/qwen-vl","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qwen-VL alternativesSmolDocling alternativesBest OCR models in 2026