Mistral OCR 4 costs $4 per 1,000 pages through the API and $2 per 1,000 pages through the Batch API. Document AI, which returns structured JSON against a schema you write, is listed at $5 per 1,000 pages. Those rates buy text and layout, not a finished spreadsheet of vendor, date, sales tax and total.
Accepts PDF, JPG, PNG, BMP, HEIC and TIFF
Upload your receipts and invoices
Drop files here or click to upload
Upload failed, please try again
Up to 50 files
Uploading...
Mistral OCR is one of the least expensive document models on the market per page, and it is genuinely good at reading messy layouts. The page rate is only the first line of the budget. Everything between a Markdown response and a clean row in your books is work somebody still has to build, run and maintain.
The base OCR endpoint returns each page as Markdown with tables, images and bounding boxes. It does not decide which number is the total or which lines are items. That logic is yours to write and test.
Document AI annotations return JSON shaped to a schema you define, with the OCR output passed to a second Mistral model. Designing, versioning and validating that schema for every receipt layout is an engineering project.
Half price applies only through the asynchronous Batch API. A bookkeeper who wants this morning of receipts in a spreadsheet before lunch pays the standard rate or waits for the job to finish.
The Mistral model page lists mistral-ocr-2505 with a May 31, 2026 retirement date and OCR 4.1 as its replacement. Pinning a version and re-testing on every migration is part of the real cost of a raw model.
Most US finance teams are not shopping for an OCR model. They want vendor, date, sales tax, total and line items in Excel or QuickBooks, checked, without a developer on call. ReceiptOCR is that finished workflow, priced per page with nothing to build.
Every receipt and invoice comes back as named columns: vendor, date, invoice number, subtotal, sales tax, tip, total and each line item. No parser to write between the model and your spreadsheet.
Download one clean file with consistent headers across every document in the batch, ready for a pivot table, a CPA or an import into your accounting system.
Check each extracted value against the image in a browser before it reaches your books, which is the control a raw API response never gives you.
Drop in hundreds of photos and PDFs at once and get results in minutes, without writing JSONL files, polling job status or waiting for an asynchronous queue.
Developers can call the same extraction over a REST API and receive parsed fields as JSON, so you can start in the browser and automate later without switching vendors.
Plans start at $49 a month for 2,500 pages with no per-seat fees, no schema to maintain and no second model call to budget for.
The same outcome a Mistral OCR integration is meant to produce, in about a minute.
Drag in phone photos, scanned PDFs and emailed invoices. One file or a few hundred in a single batch.
Tip: Run a sample of your hardest documents first: faded thermal paper, handwritten tips and multi-page vendor bills.
Vendor, date, totals, sales tax and every line item are read into consistent columns, with no prompt, schema or post-processing to write.
Check the values against the images, then download Excel or CSV or send the data to QuickBooks.
Written for US developers, product leads and finance operators comparing a raw OCR model against a finished extraction tool for receipts and invoices.
If you are building a product, Mistral OCR is a strong, cheap engine. Budget the parsing, validation and review layer on top, because that is where most of the time goes.
You need client receipts in a spreadsheet by the end of the day, not an API key. A finished tool is cheaper than the hours needed to turn Markdown into ledger rows.
Month-end volume arrives in spikes. Predictable monthly pricing and a review screen beat a per-page meter plus an internal tool nobody on the team owns.
When someone proposes adding Mistral OCR, the useful question is total cost per usable record, including engineering time, not the per-page rate on the model card.
Every Mistral rate on this page was read from Mistral announcement and documentation pages in US dollars. Where Mistral does not publish a number, the cell says so rather than guessing.
Mistral OCR 4 costs $4 per 1,000 pages through the standard API and $2 per 1,000 pages through the Batch API, which carries a 50% discount. Document AI, the structured-output layer on the same endpoint, is listed at $5 per 1,000 pages. The previous generation, Mistral OCR 3, launched at $2 per 1,000 pages, or $1 in batch, and remains available for existing integrations.
That makes Mistral one of the cheapest ways to turn a page image into structured text. It does not make it the cheapest way to turn a receipt into a bookkeeping entry, and the difference between those two things is the whole point of this page.
| Product | Model | Standard API, per 1,000 pages | Batch API, per 1,000 pages | What comes back |
|---|---|---|---|---|
| Mistral OCR 4 | mistral-ocr-latest (OCR 4.1) | $4.00 | $2.00 | Markdown text, tables, images, bounding boxes, block labels, confidence scores |
| Document AI | OCR 4 plus annotations | $5.00 | Not published | JSON shaped to your own schema |
| Mistral OCR 3 | mistral-ocr-2512 | $2.00 | $1.00 | Markdown text and tables |
| OCR 3 annotations | mistral-ocr-2512 | $3.00 | Not published | JSON shaped to your own schema |
Sources: the Mistral OCR 4 announcement of June 23, 2026 and the Mistral OCR 3 announcement of December 17, 2025. The announcements state the Document AI and annotation prices per 1,000 pages but do not spell out whether they replace or add to the base OCR rate. Until your first invoice settles it, budget the higher case: $4 plus $5, or $9 per 1,000 annotated pages on OCR 4.
Mistral OCR 4 costs $0.004 per page at the standard rate and $0.002 per page through the Batch API. Mistral OCR 3 costs $0.002 per page, or $0.001 in batch. A single-page receipt is one page, so a receipt costs four tenths of a cent to read on OCR 4, before any work to turn the text into fields.
Here is the same monthly volume priced on each Mistral option, assuming single-page receipts. The arithmetic is ours, from the published rates.
| Pages per month | OCR 4 standard | OCR 4 batch | Document AI at $5 | Document AI plus OCR, higher case |
|---|---|---|---|---|
| 1,000 | $4 | $2 | $5 | $9 |
| 2,500 | $10 | $5 | $12.50 | $22.50 |
| 10,000 | $40 | $20 | $50 | $90 |
| 100,000 | $400 | $200 | $500 | $900 |
Those are honest numbers and they are low. At 2,500 receipts a month, the model bill is somewhere between $5 and $22.50. For comparison, the ReceiptOCR Starter plan is $49 a month, or $24 a month billed annually, for 2,500 pages. On raw page cost Mistral wins clearly, and nobody should pretend otherwise.
What the Mistral figure leaves out is everything after the response arrives: choosing which number is the total when a receipt shows three, splitting sales tax from tip, mapping vendor names, handling a multi-page invoice, retrying failures, building a screen where a person can check a value against the image, and exporting to Excel or QuickBooks. If one developer spends one day a month maintaining that pipeline, the labor dwarfs a $10 model bill many times over.
No. Mistral OCR is a paid, metered API billed per 1,000 pages. The Mistral Free plan lets you test models in Studio, but neither the pricing page nor the OCR model pages list a standing monthly free page allowance for OCR, the way Google Cloud Vision includes its first 1,000 units each month. Treat any "free Mistral OCR" claim as a trial, not a tier.
No. Mistral OCR is a proprietary model served through the Mistral API and the Document AI interface in Mistral Studio. The OCR 4 announcement says self-managed deployment, running on a single container inside your own environment, is available to enterprise customers, which is a commercial arrangement rather than an open weights release.
You submit many OCR requests as one asynchronous job and pay 50% of the standard rate. The Mistral batch documentation supports the /v1/ocr endpoint, accepts batch files of up to 1 million requests in JSONL format, and allows inline batching without a file for jobs under 10,000 requests. Results are retrieved when the job completes, not in real time.
For a nightly archive of scanned documents, batch is the obvious choice and halves the bill. For an accountant clearing a client folder during the working day, the wait usually matters more than the savings on a bill that was already small.
Uploaded document files must be 50 MB or less and contain 1,000 pages or fewer, according to the Mistral OCR documentation FAQ. Supported inputs include PDF, PPTX and DOCX documents and PNG, JPEG and AVIF images, passed as a public URL, a Base64 string or an uploaded file. Larger archives have to be split before submission.
It reads them well, and Mistral says OCR 3 works well on invoices, receipts and forms. Reading is not extracting, though. The base endpoint returns the page as Markdown, so to get fields you either parse that text yourself or use Document AI annotations, where you define a JSON schema and Mistral passes the OCR output to a second model, mistral-small-2603, to fill it.
Either route still leaves validation to you. A schema tells the model what to return; it does not guarantee that the subtotal plus tax equals the total, that the date is in the right year, or that a faded digit was read correctly. Those checks are what a finished extraction product exists to do.
Compared on the same unit, Mistral sits between the hyperscalers plain text OCR and their prebuilt receipt models. It costs more than raw text detection but returns richer layout, and it costs far less than the purpose-built expense parsers, which do the field extraction for you.
| Service | What it does | Price per 1,000 single-page receipts |
|---|---|---|
| Mistral OCR 4, batch | Layout-aware text and tables | $2.00 |
| Mistral OCR 4, standard | Layout-aware text and tables | $4.00 |
| AWS Textract DetectDocumentText | Plain text OCR | $1.50 |
| Google Document AI Enterprise OCR | Plain text OCR | $1.50 |
| AWS Textract AnalyzeExpense | Receipt and invoice fields | $10.00 |
| Azure Document Intelligence prebuilt receipt | Receipt fields | $10.00 |
| Reducto Extract | Schema-based field extraction | $20.00 |
| Google Document AI Expense parser | Receipt fields, billed per document | $100.00 |
The AWS, Google and Azure figures are the list rates we verified for our AWS Textract pricing, Google Document AI pricing and Azure Document Intelligence pricing pages. The Reducto rate is from its own rate card, broken down in our Reducto pricing review. LlamaParse, which bills credits by tier, is covered on our LlamaParse pricing page. For every meter in one table, see the OCR API pricing comparison.
Buy Mistral OCR when you are a developer building document handling into a product, you have volume in the tens of thousands of pages or more, and you have engineering time to own parsing, validation and review. At that scale the page rate matters and the pipeline cost is spread across a lot of documents. It is also a sensible engine if you are moving from a chat model to a document model, a trade-off we cover on the LLM OCR page.
Skip it when the goal is simply receipts and invoices in a spreadsheet for a business, a bookkeeping practice or a finance team. In that case the model is a component, and you would be buying one part of a product and building the rest yourself. Upload a batch to the extractor at the top of this page and compare the columns with what a Mistral response gives you. That test takes a minute and settles the question on your own documents.
Mistral OCR 4 costs $4 per 1,000 pages through the standard API and $2 per 1,000 pages through the Batch API. Document AI, which returns JSON to your schema, is listed at $5 per 1,000 pages. Mistral OCR 3 remains available at $2 per 1,000 pages, or $1 in batch.
Mistral OCR 4 costs $0.004 per page at the standard rate and $0.002 per page in batch. Mistral OCR 3 costs $0.002 per page, or $0.001 in batch. A single-page receipt counts as one page, so reading one receipt on OCR 4 costs less than half a cent before any field extraction.
No. Mistral OCR is a metered API billed per 1,000 pages. The Mistral Free plan lets you test models in Studio, but Mistral does not list a standing monthly free page allowance for OCR on its pricing or model pages, unlike Google Cloud Vision, which includes 1,000 units a month.
No. Mistral OCR is a proprietary model available through the Mistral API and Document AI in Mistral Studio. Self-managed deployment on your own infrastructure is offered to enterprise customers as a commercial arrangement, not as an open weights download.
The Batch API charges 50% of the standard rate, which takes Mistral OCR 4 from $4 to $2 per 1,000 pages and OCR 3 from $2 to $1. Batch jobs are asynchronous, support up to 1 million requests per file, and allow inline submission for jobs under 10,000 requests.
Uploaded document files must be 50 MB or less and contain 1,000 pages or fewer, according to the Mistral OCR documentation. Split larger PDFs before submitting them. Inputs can be PDF, PPTX, DOCX or images such as PNG, JPEG and AVIF.
Yes, it reads them well, but the base endpoint returns Markdown rather than fields. To get vendor, date, tax and total as data you either parse the text yourself or use Document AI annotations with your own JSON schema, and you still need to validate the values before they reach your books.
It depends which Textract meter you compare. Textract plain text detection is $1.50 per 1,000 pages, cheaper than Mistral OCR 4 at $4. Textract AnalyzeExpense, which returns receipt fields, is $10 per 1,000 pages, so Mistral is cheaper for reading but leaves the field extraction to you.
If you want receipt and invoice fields in Excel, CSV or QuickBooks without building a pipeline, a finished extraction tool is the practical alternative. ReceiptOCR returns vendor, date, sales tax, total and line items with a review screen, from $49 a month for 2,500 pages.
Stop typing receipts by hand
Upload your receipts and invoices and get a clean Excel or CSV file in minutes.
Extract my receipts nowFree to try, no sign up required
When a language model beats a classic OCR engine, and when it does not.
Every published OCR meter in one per 1,000 page table.
Enterprise OCR, the Expense parser and the per document trap.
Read, prebuilt and custom model rates with the commitment ladder.
Parsed receipt fields as JSON, without writing layout heuristics.
Invoice headers and line items over a REST API.
Get started extracting receipt and invoice data to spreadsheets.
USD
per month
per month
$288 charged today
Choose speed vs accuracy when extracting
| Base AI Faster | 2,500 pages |
| Pro AI Best accuracy | 500 pages |
Scale receipt and invoice extraction across your team with automation.
USD
per month
per month
$888 charged today
Choose speed vs accuracy when extracting
| Base AI Faster | 10,000 pages |
| Pro AI Best accuracy | 2,000 pages |
Enterprise‑grade receipt and invoice extraction and controls.
USD
per month
per month
$ charged today
Choose speed vs accuracy when extracting
| Base AI Faster | pages |
| Pro AI Best accuracy | pages |
Custom workflows and deployment for AI automation at scale