On 23 June 2026, Mistral AI announced Mistral OCR 4. The fourth OCR generation moved beyond markdown extraction into localized, typed and confidence-scored document structure.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 23 June 2026 |
| Availability or weight release | 23 June 2026 through the API and Document AI, with enterprise self-hosting |
| Release type | specialist document-understanding model |
| Access | hosted API, Mistral Document AI, partner platforms and single-container self-hosting for enterprise customers |
| Architecture | a compact multimodal document parser returning text, layout, block type and calibrated confidence metadata |
| Maximum stated context | document and page limits set by the OCR API rather than a chat-style token window |
What changed
OCR 4 turns document ingestion into an auditable data product. Bounding boxes allow citations and redactions to point back to a region, block types separate equations or tables from prose, and per-word confidence can route uncertain fields to review. Mistral also published benchmark limitations instead of treating aggregate string matching as ground truth, which is important for financial, legal and scientific documents where a benchmark reference can itself be wrong.
The practical comparison is therefore not simply whether Mistral OCR 4 has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| OlmOCRBench | 85.20 | Top reported aggregate, with documented scoring caveats |
| OmniDocBench | 93.07 | Release score that Mistral explicitly asks readers to qualify |
| Human preference | 72% average win rate across 600+ documents | Blind comparison spanning more than 12 languages |
These are release-time results, not independently reproduced guarantees. Mistral identifies scoring artifacts in OlmOCRBench and OmniDocBench; human-preference results cover 600-plus documents but remain vendor-run. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
Mistral OCR 4 is described as a compact multimodal document parser returning text, layout, block type and calibrated confidence metadata with document and page limits set by the OCR API rather than a chat-style token window of stated context. Its access position at verification time is hosted API, Mistral Document AI, partner platforms and single-container self-hosting for enterprise customers. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For Mistral OCR 4, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
OCR 4 is a major specialist release because it improves both extraction and evidence traceability. Buyers should test field-level recall and confidence calibration on their own scans before automating downstream actions.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: Mistral AI official release artwork. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.



