Mistral launches OCR 4 with structured output, self-hosted deployment, and Microsoft Foundry availability
What happened
France's Mistral AI launched OCR 4, a document model targeted at regulated industries (finance, legal, healthcare) that need to keep deployments on-prem.
Context and impact
Mistral is attacking the gap left by Anthropic and OpenAI's cloud-only posture. The self-hosted option and day-one Microsoft Foundry availability are clear enterprise signals. The launch also fits the broader European sovereign AI push (Apertus, the EUROPA consortium).
Details
- 170 languages with structure awareness — headings, tables, signatures
- Bounding boxes + block classification + confidence scores in JSON
- Pricing: $4 per 1,000 pages
- Available via Mistral API, Amazon SageMaker, Microsoft Foundry (day one)
- Self-hosted option for regulated customers
Open original source
VentureBeat