Mistral OCR vs docTR
Pick Mistral OCR when you need scanned PDFs turned into Markdown with tables for RAG, and you are happy to call a hosted API at $2 per 1,000 pages. Pick docTR when you need word-level text with coordinates, want to run it yourself under Apache-2.0, or want to train your own models. docTR returns only words and lines, so it cannot give you Markdown or table structure.
Mistral OCRmistral/mistral-ocr | docTRmindee/doctr | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 25 credits / page | 1 credit / page |
| Per 1,000 pages | $25.00 | $1.00 |
| Typical speed | 0.9 s / page | 0.9 s / page |
| Best for | High-volume PDF to Markdown | Word-level OCR with positions |
| Inputs | PDF, images | PDF, images |
| Languages | 100+ languages | Mainly Latin scripts |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | ✓ Yes |
| 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 Mistral OCR if
- You need PDF to Markdown output, and OCR 3 rebuilds tables as HTML with rowspan and colspan.
- You process pages in bulk: $2 per 1,000 pages, or $1 per 1,000 through the Batch API.
- You do not want to run GPUs, since it is a hosted API that returns Markdown, HTML, JSON and images.
Choose docTR if
- You need word, line and block structure with coordinates as JSON, with straight or rotated boxes.
- You must self-host: it is Apache-2.0, runs on CPU or NVIDIA GPU, and costs only your compute.
- You feed invoices or receipts into your own extraction, or your Python ML team wants to train its own models and pick detection and recognition architectures.
Key differences
| Mistral OCR | docTR | |
|---|---|---|
| How you run it | Hosted API (current version mistral-ocr-2512, OCR 3) | Python library on PyTorch, with Docker images; GPU images need CUDA 12.2 or newer on the host |
| Licence | Proprietary | Apache-2.0 |
| Price | $2 per 1,000 pages; $3 with structured annotations; $1 per 1,000 with Batch API | Free, self-hosted (you pay for compute) |
| Output formats | Markdown, HTML, JSON, images | JSON, plain text, bounding boxes |
| Tables | OCR 3 reconstructs tables as HTML with rowspan and colspan | No table, formula or Markdown output; words and lines only |
| Benchmark | OmniDocBench v1.6 overall: 85.66; the first version scored 78.83 on v1.5 | Not published |
| Known failure | Hallucinated text on unreadable handwriting; changed checkbox states and misread characters on a dense financial table page; colored-background pages returned as images in the 2025 release | Cropping images gives poorer results (issue #2007); split-crop merging misjoins repeated characters (issue #2229); ONNX export of MASTER truncates decoding steps (issue #2224) |
Sources
- Mistral OCR 3 announcement
- VentureBeat on OCR 3 pricing
- OmniDocBench v1.6 leaderboard
- Reducto test of Mistral OCR
- docTR GitHub README
- Issue #2229
What each one is
Mistral OCR
Fast document OCR that returns clean Markdown with tables and embedded images.
Mistral OCR detailsdocTR
Open-source OCR library from Mindee with separate text detection and recognition models.
docTR 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": "mistral/mistral-ocr","fallbacks": ["mindee/doctr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Mistral OCR alternativesdocTR alternativesBest OCR models in 2026