Mistral OCR vs Nanonets-OCR
Pick Mistral OCR for high-volume PDF to Markdown and scanned business documents through a hosted API, where the $2 per 1,000 pages price ($1 via Batch API) and HTML table output matter most. Pick Nanonets-OCR if you must self-host and want signatures, watermarks, checkboxes and flowcharts tagged in the output. Neither is safe on handwriting, so review outputs when figures must be exact.
Mistral OCRmistral/mistral-ocr | Nanonets-OCRnanonets/nanonets-ocr | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 25 credits / page | 2 credits / page |
| Per 1,000 pages | $25.00 | $2.00 |
| Typical speed | 0.9 s / page | 1.8 s / page |
| Best for | High-volume PDF to Markdown | Markdown with semantic tags |
| Inputs | PDF, images | PDF, images |
| Languages | 100+ languages | Multilingual |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | No | ✓ Yes |
| 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 Mistral OCR if
- You want a hosted API with no GPU to run, priced at $2 per 1,000 pages, or $1 per 1,000 through the Batch API.
- You need tables rebuilt as HTML with rowspan and colspan, which OCR 3 does.
- You need stronger measured accuracy: 85.66 overall on OmniDocBench v1.6, scored independently.
Choose Nanonets-OCR if
- You need to self-host the weights from Hugging Face and pay only for compute.
- Your documents carry signatures, watermarks or checkboxes, which it tags or converts to unicode symbols.
- You convert flowcharts and org charts to Mermaid code, or need inline and display LaTeX kept distinct.
Key differences
| Mistral OCR | Nanonets-OCR | |
|---|---|---|
| How you run it | Hosted API from Mistral AI, pinned version mistral-ocr-2512 (OCR 3). | Self-hosted 3B model (Hugging Face lists about 4B params), fine-tuned from Qwen2.5-VL-3B. |
| Licence | Proprietary. | Not stated on the Nanonets-OCR2-3B model card (no licence tag). The base model is Qwen2.5-VL-3B. |
| Price | $2 per 1,000 pages; $3 with structured annotations; $1 per 1,000 with the Batch API. | Free software. You pay for your own compute. |
| Output formats | Markdown, HTML, JSON and extracted images. | Markdown, HTML, LaTeX and Mermaid. |
| Benchmark | OmniDocBench v1.5 overall 78.83 for the first version (OCR 2503); v1.6 overall 85.66. Both independent. | olmOCR-bench overall 69.5, DocVQA 89.43, ChartQA 78.56 for OCR2-3B. All reported by the authors. |
| Handwriting | Hallucinated text reported on unreadable handwriting. Mistral reports a 74% win rate for OCR 3 over OCR 2 on forms, scans, tables and handwriting. | Not trained on handwriting and can hallucinate. One handwriting test saw generation loops with CER up to 2100% and 69.1s per image on an L4. |
| Known failure | Colored-background pages returned as images with no OCR text (2025 release). Reducto found changed checkbox states and misread characters on a dense financial table. | Trained on a fixed prompt, so hallucination with other prompts is expected. Loses to Gemini 2.5 Flash in 52.43% of head-to-head Markdown comparisons, wins 39.98%. |
Sources
- Mistral OCR 3 announcement
- OmniDocBench v1.6 leaderboard
- Reducto test of Mistral OCR
- Nanonets-OCR2-3B model card
- HF discussion fixed prompt
- Nanonets-OCR-s announcement
What each one is
Mistral OCR
Fast document OCR that returns clean Markdown with tables and embedded images.
Mistral OCR detailsNanonets-OCR
Open image-to-Markdown model that tags tables, equations, signatures, watermarks and checkboxes.
Nanonets-OCR 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": ["nanonets/nanonets-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Mistral OCR alternativesNanonets-OCR alternativesBest OCR models in 2026