Granite Docling vs Pixtral
Pick Granite Docling for converting English PDFs with tables, code and equations inside a Docling pipeline, where it is small (258M) and cheap to run. Pick Pixtral only if you already use the Mistral API or want question answering over charts and documents. Pixtral is deprecated and reported to miss OCR on dense or non-Latin text, so it is a poor choice for a new OCR project.
Granite Doclingibm/granite-docling | Pixtralmistral/pixtral | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.5 s / page | 3.0 s / page |
| Best for | Small-model document conversion | Questions about document images |
| Inputs | PDF, images | images |
| Languages | English, experimental CJK and Arabic | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| 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 Granite Docling if
- You need structured conversion: it outputs DocTags, Markdown, HTML, LaTeX and bounding boxes for the Docling library.
- Your documents are English PDFs with tables, code or equations: table structure TEDS is 0.97 on FinTabNet, equation F1 is 0.968 and code F1 is 0.988 per the model card.
- You want low-cost self-hosting on CPU or a small GPU, since the model has only 258M parameters under Apache-2.0.
Choose Pixtral if
- You need question answering over documents and charts: it scores 90.7 ANLS on DocVQA and 81.8 on ChartQA, as reported by Mistral.
- You already have an integration on the Mistral API, which offered it as pixtral-12b-2409 at a launch price of $0.15 per million input and output tokens.
- You need to send many images in one request, because the context window is 128K tokens.
Key differences
| Granite Docling | Pixtral | |
|---|---|---|
| Primary job | Document conversion to DocTags for Docling | General vision language model for question answering over documents and charts |
| Size | 258M parameters | 12.4B parameters (12B decoder + 400M vision encoder) |
| Licence | Apache-2.0 | Apache-2.0 open weights |
| How you run it | Self-host only, typically inside Docling | Mistral API or self-host open weights |
| Price | Free, you pay for compute | $0.15 per million input and output tokens at API launch; free to self-host |
| Output formats | DocTags, Markdown, HTML, LaTeX, bounding boxes | Plain text, Markdown, JSON |
| Known failure | Repetitive loops on page headings and invoices; exclamation-mark output on GPUs without bfloat16 such as T4 unless run in float32; Japanese, Arabic and Chinese are experimental | Deprecated by Mistral and no longer maintained; reported accuracy drop on handwriting, small fonts and non-Latin scripts; Q4 quantization degrades small text OCR |
Sources
- Granite-Docling model card
- HF discussion #24 repetition
- Ollama issue #12355
- Mistral announcement
- Mistral Sept 2024 pricing
- HN launch thread
What each one is
Granite Docling
Compact IBM vision model that converts pages to DocTags for the Docling pipeline.
Granite Docling detailsPixtral
Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.
Pixtral 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": "ibm/granite-docling","fallbacks": ["mistral/pixtral"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Granite Docling alternativesPixtral alternativesBest OCR models in 2026