Mistral OCR 4.1
Mistral's current OCR model, adding paragraph-level bounding boxes and block confidence scores
Context window
Not published
Input / 1M tokens
$4.00 per 1,000 pages
Output / 1M tokens
$5.00 per 1,000 annotated pages
Provider
Mistral AI
Billed per page, not per token: $4 per 1,000 pages, and $5 per 1,000 annotated pages when you use Document AI structured annotations. The same price as OCR 4. · Data verified 2026-11-11
Mistral OCR 4.1 (API ID mistral-ocr-4-1) is Mistral AI's current document OCR model and the engine of its Document AI stack. Mistral released it on July 16, 2026 and made it generally available on August 31, 2026. It adds native paragraph-level bounding boxes, structural block labels and block-level confidence scores to OCR 4's Markdown output: with include_blocks enabled, each page returns a blocks array in reading order. It replaced OCR 4, which Mistral retired on September 30, 2026, and the mistral-ocr-latest alias now points to it.
How to access it
Mistral API OCR endpoint (/v1/ocr) with model ID mistral-ocr-4-1 or the mistral-ocr-latest alias, using the official mistralai SDK; large jobs can go through the Batch API (/v1/batch). Generally available since August 31, 2026.
Strengths
- ✓Paragraph-level bounding boxes with structural block labels
- ✓Block-level confidence scores for flagging uncertain regions
- ✓Markdown output with optional HTML tables and structured JSON annotations
- ✓Per-page pricing at $4 per 1,000 pages
Best for developers who...
When to choose it (and when not to)
Reach for Mistral OCR 4.1 when...
- →You need text plus element positions from PDFs, scans or office files
- →You want per-block confidence scores to route uncertain content to review
- →You already use OCR 4: calls to mistral-ocr-latest now reach OCR 4.1 at the same price
Look elsewhere if...
- ✕You need a general chat or reasoning model (this is OCR only)
- ✕Your input is plain text with no layout to extract
How to use it
- ›Set include_blocks=True to get paragraph-level boxes and block labels
- ›Pass confidence_scores_granularity='block' to get a confidence score per block
- ›Request table_format='html' when tables must keep their structure
Quickstart
Pythonimport os
from mistralai.client import Mistral
client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])
ocr_response = client.ocr.process(
model="mistral-ocr-4-1",
document={
"type": "document_url",
"document_url": "https://arxiv.org/pdf/2201.04234",
},
include_blocks=True,
confidence_scores_granularity="block",
)
print(ocr_response.pages[0].markdown)Uses the official mistralai SDK (pip install mistralai). Based on Mistral's basic OCR guide; billing is per page processed.
API model id: mistral-ocr-4-1
Compare Mistral OCR 4.1
Compare Mistral OCR 4.1 with any other model
Build a comparison →All model comparisons →