Mistral OCR 4 API vs Document AI: Choosing an Integration

Mistral OCR 4 API vs Document AI: Choosing an Integration

Mistral OCR 4 was announced on June 23, 2026 for structured document extraction. Mistral describes one OCR 4 API endpoint and a Document AI layer built on the same engine. The choice is therefore about the output and workflow a team needs, not about switching to a different OCR model.

Start with extraction when raw document structure matters

Mistral says each OCR 4 request returns extracted content, bounding boxes, block types, confidence scores and markdown-structured text. That makes direct extraction relevant when an application needs to work with raw OCR output and apply its own downstream logic.

The announcement positions this route for document ingestion, search, retrieval pipelines and high-volume or batch processing. OCR 4 supports 170 languages across 10 language groups and accepts enterprise formats including PDF, DOC, PPT and OpenDocument.

Add Document AI for shaped output

According to Mistral, Document AI uses the same endpoint with additional parameters. A team can pass a JSON schema to receive content shaped to its specification, supply an image-annotation schema or add a custom prompt to guide interpretation or summarisation of extracted document content.

This is useful when a workflow needs fields in a known structure rather than raw extracted blocks. It does not remove the need to validate results in the context where they will be used.

Pricing stated in the announcement

Mistral lists OCR 4 API pricing at $4 per 1,000 pages. It states that the Batch API discount reduces this to $2 per 1,000 pages. Document AI is listed at $5 per 1,000 pages. Teams should recheck Mistral’s current pricing before integrating a production workflow.

Where self-hosting fits

Mistral also describes OCR 4 as deployable in a single container, with a fully self-hosted option for data residency, sovereignty and compliance needs; self-managed deployment is stated to be available to enterprise customers. The announcement does not provide an Ollama workflow, downloadable-weight instructions or consumer hardware requirements, so those details should not be inferred from it.

Limits and evaluation

Mistral says OCR 4 is not a decision-maker. It is not intended for medical diagnosis, legal advice or judgment, high-stakes financial decisions, safety-critical systems, real-time or latency-sensitive processing, or non-document inputs. Mistral also advises evaluating on representative documents because aggregate benchmarks are directional.

Official source

• Mistral AI — Mistral OCR 4: SOTA OCR for Document Intelligence