Qwen-VL vs Unstructured
Pick Qwen-VL when you want one open-weights model that reads photos and screenshots and also answers questions about them, and you can add loop checks. Pick Unstructured when you need to ingest many file types (PDF, Word, PowerPoint, email, HTML) into typed elements for RAG. Qwen-VL is a vision-language model that you prompt; Unstructured is a document pipeline that partitions files.
Qwen-VLqwen/qwen-vl | Unstructuredunstructured/unstructured | |
|---|---|---|
| Type | Open source | OCR API |
| Price | 2 credits / page | 20 credits / page |
| Per 1,000 pages | $2.00 | $20.00 |
| Typical speed | 2.0 s / page | 3.0 s / page |
| Best for | Flexible extraction with prompts | Mixed file types into RAG |
| Inputs | PDF, images | PDF, images, DOCX |
| Languages | 30+ languages | Multilingual |
| Max pages per request | 200 | 1,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | 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 Qwen-VL if
- Choose it for OCR plus reasoning or VQA in one model, since it is a general vision-language model.
- Choose it for photos and screenshots with text, including blur, low light and tilted text per Qwen.
- Choose it for a wide size range (2B to 235B-A22B MoE) under Apache-2.0, with OCR in 32 languages.
Choose Unstructured if
- Choose it to ingest PDFs, HTML, Word, PowerPoint, emails and images through one library.
- Choose it for born-digital PDFs with the fast strategy, or to trade speed for quality with Auto, Fast, High-Res and VLM strategies.
- Choose it to start on the hosted API with 10,000 free pages, then $0.015 per page.
Key differences
| Qwen-VL | Unstructured | |
|---|---|---|
| What it is | General vision-language model family (latest Qwen3-VL) with OCR and document parsing | Document ETL library and platform that partitions files into typed elements |
| Licence | Apache-2.0 (Qwen3-VL repository) | Apache-2.0 for the open-source library; proprietary for the hosted platform |
| Price | Free self-hosted (you pay for compute). Hosted pricing on Alibaba Cloud Model Studio: not published here | Hosted: first 10,000 pages free, then $0.015 per page. Open-source library is free |
| Output formats | Plain text, HTML, bounding boxes, JSON, Markdown | JSON, HTML, plain text, bounding boxes |
| Benchmark | Qwen 3 VL 8B scores 64.6 on olmOCR-bench in an independent table, below dedicated OCR models | Table cell content: 0.426 open-source vs 0.866 hosted (authors' own benchmark) |
| Hardware | FP8 checkpoints need an H100-class GPU and CUDA 12+; the 235B example uses 8-way tensor parallel on vLLM | Not published |
| Known failure | Can loop and repeat the same text, mostly on low-quality images and worse with greedy decoding; GGUF builds of Qwen2.5-VL-32B often loop on scanned documents | hi_res tables can cut off at the edges and drop first letters; table OCR ignores language settings; usage telemetry is sent by default |
Sources
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- Chandra README benchmark table
- Unstructured pricing
- Unstructured GitHub README
- Issue #3262
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 detailsUnstructured
Document ETL for RAG. Partitions many file types into typed elements; open-source library too.
Unstructured 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": ["unstructured/unstructured"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qwen-VL alternativesUnstructured alternativesBest OCR models in 2026