DeepSeek-VL2 vs Chunkr
Pick DeepSeek-VL2 if you want to self-host an open-weights model for question answering on single-page documents and charts. Pick Chunkr if you need to turn PDFs, slides, Word files and images into HTML, Markdown or JSON with bounding boxes for RAG, through a cloud API or a self-hosted stack. The main split is that DeepSeek-VL2 is a vision language model, while Chunkr is a document parsing service with layout analysis and chunking.
DeepSeek-VL2deepseek/deepseek-vl2 | Chunkrchunkr/chunkr | |
|---|---|---|
| Type | Vision model | OCR API |
| Price | 2 credits / page | 14 credits / page |
| Per 1,000 pages | $2.00 | $14.00 |
| Typical speed | 3.0 s / page | 3.0 s / page |
| Best for | Charts and tables | Layout-aware chunks for RAG |
| Inputs | images | PDF, images, DOCX |
| Languages | Mainly English and Chinese | Multilingual |
| Max pages per request | 500 | 1,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | ✓ 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 DeepSeek-VL2 if
- Its full model scores 93.3 on DocVQA test, so it suits question answering over single-page documents.
- The weights allow commercial use under the DeepSeek Model License, and the code is MIT.
- Only 1.0B to 4.5B parameters are active per token, and Tiny runs under 40GB of GPU memory.
Choose Chunkr if
- It returns HTML, Markdown, JSON and OCR with bounding boxes, and it chunks output for RAG.
- It accepts PDF, PPT, Word and images, so you do not need to convert slides or Word files first.
- You can use the hosted API or self-host with Docker (GPU, CPU-only and Mac ARM options) against any OpenAI-compatible LLM endpoint.
Key differences
| DeepSeek-VL2 | Chunkr | |
|---|---|---|
| What it is | Mixture-of-experts vision language model in Tiny, Small and full sizes | Document parsing service with layout analysis, OCR and semantic chunking |
| How you run it | Self-host the open weights only | Cloud API or open-source self-host; self-host needs an external LLM endpoint |
| Licence | MIT (code); DeepSeek Model License (weights, commercial use allowed) | AGPL-3.0 or commercial license; closed-source products need the commercial license |
| Price | Free, you pay for compute | Hosted credit-based plans, official prices not verified; open-source version is free, you pay for compute and an LLM endpoint |
| Output formats | Plain text, Markdown | HTML, Markdown, JSON, bounding boxes |
| Benchmarks | DocVQA test 93.3, ChartQA test 86.0, OCRBench 834 (Small), all from the authors' paper | Not published |
| Known limits | Only a few images per session, dynamic tiling turns off above two images, struggles with blurry images, Small needs about 80GB GPU memory without chunked prefill | Open-source version uses community models that are less accurate than the cloud's in-house models, and it does not support Excel |
Sources
What each one is
DeepSeek-VL2
DeepSeek's mixture-of-experts vision models (December 2024) with OCR and table and chart understanding.
DeepSeek-VL2 detailsChunkr
Document parsing API that segments and chunks pages for RAG. Self-hostable version available.
Chunkr 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": "deepseek/deepseek-vl2","fallbacks": ["chunkr/chunkr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
DeepSeek-VL2 alternativesChunkr alternativesBest OCR models in 2026