Pixtral vs Phi-4-multimodal
Pick Phi-4-multimodal for English document Q&A on modest GPUs, since it scores DocVQA 93.2 at 5.6B parameters under an MIT licence. Pick Pixtral only if you already run on the Mistral API, because Mistral marks Pixtral 12B as deprecated and no longer maintained. Neither suits non-English or handwritten transcription.
Pixtralmistral/pixtral | Phi-4-multimodalmicrosoft/phi-4-multimodal | |
|---|---|---|
| Type | Vision model | Vision model |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 3.0 s / page | 2.5 s / page |
| Best for | Questions about document images | Small model for OCR and questions |
| Inputs | images | images |
| Languages | Multilingual | English for vision |
| Max pages per request | 500 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | No | No |
| Confidence | No | No |
| 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 Pixtral if
- You already have an integration on the Mistral API, where it launched at $0.15 per million input and output tokens.
- You want Apache-2.0 open weights and a 128K token context that fits many images per request.
- Your job is question answering over charts and documents, where Mistral reports DocVQA 90.7 ANLS and ChartQA 81.8.
Choose Phi-4-multimodal if
- You work with English documents and want a higher DocVQA score (93.2) from a smaller 5.6B model.
- You need audio and images handled in one model, for example a mixed audio and image pipeline.
- You want the MIT licence and free self-hosting, with Azure AI Foundry as another way to run it.
Key differences
| Pixtral | Phi-4-multimodal | |
|---|---|---|
| Licence | Apache-2.0 | MIT |
| Size | 12.4B (12B decoder + 400M vision encoder) | 5.6B |
| How you run it | Mistral API (pixtral-12b-2409) or open weights | Self-host or Azure AI Foundry |
| Price | $0.15 per million input and output tokens at API launch; free to self-host | Free, self-hosted (you pay for compute) |
| DocVQA | 90.7 ANLS (Mistral) | 93.2 (Microsoft) |
| Input types | Images and text | Text, image and audio |
| Known failure | Reported drops on handwriting, small fonts and non-Latin scripts; Q4 quantization degrades small text OCR | Vision is English only, with weaker results in other languages; the model card warns of repetitive output in long chats |
Sources
- Mistral announcement
- Mistral Sept 2024 pricing
- HN launch thread
- Third-party deployment guide
- Hugging Face model card
What each one is
Pixtral
Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.
Pixtral detailsPhi-4-multimodal
Microsoft's 5.6B text, vision and audio model (February 2025). Scores 84.4 on OCRBench per its model card.
Phi-4-multimodal 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/pixtral","fallbacks": ["microsoft/phi-4-multimodal"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Pixtral alternativesPhi-4-multimodal alternativesBest OCR models in 2026