Mistral OCR 4 costs $4 per 1,000 pages through the API, or $2 per 1,000 pages if you send work through the Batch API at the 50 percent discount. Mistral Document AI, the annotation layer that returns structured fields instead of plain text, is $5 per 1,000 pages. That is roughly half a cent a page, which is genuinely cheap. The catch for invoice work is what the rate does not include: OCR returns markdown, so turning that into vendor names, dates, line items, and a spreadsheet is code you write and maintain. This page shows the real Mistral OCR pricing, the per-page math against AWS Textract and a flat invoice tool, and how to decide which one is actually cheaper for you.
Upload your invoices
Drop files here or click to upload
Up to 50 files
Uploading...
Mistral OCR is priced well and benchmarks strongly. Before you budget for it as an invoice pipeline, it helps to know which costs sit outside the per-page figure.
The OCR endpoint gives you the page as structured markdown with tables and headings preserved. It does not hand you a vendor name, an invoice number, a due date, or a totals column. Mapping markdown into invoice fields is your code.
To get typed fields back you use the Annotations feature, billed as Document AI at $5 per 1,000 pages rather than $4. You also define the schema and keep it working as vendor layouts change.
A per-page API returns JSON to your application. Storage, review screens for low-confidence fields, approvals, and an Excel or CSV export are all things you build around it.
At $4 per 1,000 pages, 2,000 invoice pages a month costs $8. A developer spending two days wiring up parsing, retries, and exports costs far more than a year of that page volume.
Confidence scores tell you a field looks shaky. Deciding what happens next, queueing it, flagging it, routing it to a human, is workflow you design rather than something the API price includes.
Per-page billing means a heavy month costs more than a quiet one. That is fine at small scale, but it makes the line item harder to forecast than a fixed subscription.
These are the published Mistral rates as of July 2026, taken from the official Mistral API pricing page and the Mistral OCR 4 announcement. Confirm current figures with Mistral before you commit, since model pricing changes.
The standard rate for Mistral OCR 4, model ID mistral-ocr-4-0, is $4 per 1,000 pages. That works out to $0.004 a page for text and layout extraction.
Sending documents through the Batch API applies a 50 percent discount, bringing OCR to $2 per 1,000 pages. Batch trades immediate responses for the lower rate, which suits overnight invoice runs.
Document AI is the annotation layer that returns structured fields against a schema you define, rather than raw markdown. It is billed at $5 per 1,000 pages.
Mistral OCR 4 supports 170 languages across 10 language groups, which matters if you handle invoices from international suppliers.
Mistral reports a score of 85.20 on the public OlmOCRBench benchmark and human evaluator win rates averaging 72 percent against competing systems.
The model ships as a single container for self-hosted deployment, which is the route to take when data residency rules stop you sending invoices to a public API.
The per-page rate is easy. The honest comparison needs two more numbers.
Count invoice pages rather than invoices, since a three-page invoice bills as three pages. Multiply by $0.004 for the OCR API, $0.002 for batch, or $0.005 for Document AI.
Tip: At 2,000 pages a month the API cost is about $8. The page rate is almost never the deciding number.
Estimate the developer days to turn markdown or annotations into validated invoice records, plus the ongoing time to fix layouts that break. Price that at a real hourly rate and add it in.
Put the total next to a flat plan that already includes review, validation, and Excel or CSV export. If you need an API rather than an app, compare like for like against a flat-rate invoice API instead.
A raw OCR API and a finished invoice tool are not really competing products. Here is the honest split.
If you have developers and want full control over how documents flow through your own system, the Mistral rate is excellent value and the build is the point.
At hundreds of thousands of pages, $2 to $4 per 1,000 through batch is hard to beat, and the engineering cost amortizes across the volume.
The self-hosted container is a genuine advantage when invoices cannot leave your infrastructure, which most hosted invoice tools cannot offer.
If the goal is invoices in Excel this week and nobody is going to write code, a flat-rate application is cheaper in practice even though its per-page rate looks higher.
Last updated July 2026. Figures come from Mistral's published API pricing and the Mistral OCR 4 announcement of 23 June 2026. Model pricing changes, so confirm the current rate with Mistral before you commit.
Mistral prices document processing per page, with no monthly minimum and no seat fees. There are three rates that matter for invoice work.
| Product | What it returns | Price per 1,000 pages | Per page |
|---|---|---|---|
| Mistral OCR 4 (API) | Markdown text with layout, tables, bounding boxes, confidence scores | $4 | $0.004 |
| Mistral OCR 4 (Batch API) | Same output, asynchronous, 50 percent discount | $2 | $0.002 |
| Mistral Document AI | Structured fields against a schema you define (Annotations) | $5 | $0.005 |
Mistral OCR 4, released on 23 June 2026 under the model ID mistral-ocr-4-0, reads PDFs, DOCX, PPTX, PNG, JPEG, and AVIF files and returns each page as markdown that preserves the document's structure. Alongside the text you get paragraph-level bounding boxes, typed-block classification that labels titles, tables, equations, and signatures, and inline confidence scores. Mistral reports 170 languages across 10 language groups, a score of 85.20 on the public OlmOCRBench benchmark, and human evaluator win rates averaging 72 percent against competing systems. It can also be deployed as a single self-hosted container.
For a document search index or a retrieval pipeline, that is a complete answer. For accounts payable it is the first half. Markdown that faithfully reproduces an invoice table is not the same as knowing which number is the subtotal, which is the tax, and which line items belong to purchase order 4501. That mapping is the work.
Here is the honest per-page comparison against the other options teams usually shortlist. The important column is the last one, because it explains why the cheapest rate is not automatically the cheapest project.
| Option | Cost per 1,000 pages | Returns invoice fields? | What you still build |
|---|---|---|---|
| Mistral OCR 4 | $4 ($2 batch) | No, markdown text | Field mapping, validation, storage, review, export |
| Mistral Document AI | $5 | Yes, against your schema | Schema design, validation, storage, review, export |
| AWS Textract, Detect Document Text | $1.50 | No, raw text | Everything above, plus AWS plumbing |
| AWS Textract, AnalyzeExpense | $10 | Yes, invoice and receipt fields | Validation, storage, review, export |
| InvoiceExtractor Starter | ~$19.60 (flat $49/mo, 2,500 pages) | Yes, including line items | Nothing, exports to Excel, CSV, JSON, QBO |
| InvoiceExtractor Plus | ~$14.90 (flat $149/mo, 10,000 pages) | Yes, including line items | Nothing, exports to Excel, CSV, JSON, QBO |
AWS Textract figures are the published US West (Oregon) rates for the first one million pages a month. Azure AI Document Intelligence is a reasonable fourth option, but Microsoft publishes its per-page rates through the Azure pricing calculator by region rather than as a flat table, so check the Azure Document Intelligence pricing breakdown for that comparison, and the AWS Textract pricing page for the full Textract math.
For plain text extraction, no: Textract's Detect Document Text runs $1.50 per 1,000 pages against Mistral's $4, though Mistral's batch rate of $2 narrows that to almost nothing. For invoices specifically, Mistral is much cheaper. Textract's AnalyzeExpense, the endpoint that actually returns invoice and receipt fields, costs $10 per 1,000 pages, while Mistral Document AI returns structured fields for $5. So the ranking flips depending on which endpoint does the job you need.
Run the numbers on a typical mid-sized AP workload. At 2,000 invoice pages a month, Mistral OCR costs $8, or $4 on batch. Over a year that is under $100. It is close enough to free that price should not drive the decision at all.
What does drive it is the work around the API. Someone has to parse markdown into fields or design and maintain an annotation schema, reconcile line items against totals, decide what happens when confidence drops on a tax amount, store the results, build a screen where a human fixes the ones that came out wrong, and produce the Excel file the accountant actually wanted. Two days of developer time at a normal loaded rate costs more than a decade of the page fees. If your team enjoys owning that pipeline and has reasons to, the economics are outstanding. If your goal is invoices in a spreadsheet by Friday, paying roughly $15 to $20 per 1,000 pages for a tool that already does all of it is the cheaper answer, and it is why the two products rarely compete for the same buyer.
These are not mutually exclusive. Plenty of teams use a raw OCR API for bulk document text, archives, contracts, and search indexing, and a purpose-built tool for the accounts payable stack where structured fields, validation, and accounting exports matter. InvoiceExtractor sits in the second slot: upload a PDF, scan, or photo, and AI returns the vendor, dates, totals, tax, and full invoice line items ready to export. If you would rather stay in code, the flat-rate invoice data extraction API returns the same structured JSON without per-page metering. For the wider shortlist, the best invoice data extraction software comparison covers where each tool fits, and the guide to the cost to process an invoice puts any software price next to the labor it replaces.
Mistral OCR 4 costs $4 per 1,000 pages through the standard API, which is $0.004 a page. Sending documents through the Batch API applies a 50 percent discount and brings it to $2 per 1,000 pages. Mistral Document AI, which returns structured fields rather than markdown text, is $5 per 1,000 pages.
The standard API rate works out to $0.004 per page, the Batch API rate to $0.002 per page, and Mistral Document AI to $0.005 per page. There is no monthly minimum and no per-seat fee, so a month with 500 pages costs about $2 on the standard OCR rate.
Mistral OCR at $4 per 1,000 pages returns the page as markdown with layout, bounding boxes, and confidence scores. Mistral Document AI at $5 per 1,000 pages uses the Annotations feature to return typed, structured fields against a schema you define. You pay the extra dollar to skip writing the code that turns text into fields.
Mistral offers free experimentation limits on its platform for testing, but ongoing document processing is billed at the published per-page rates. Because OCR starts at $4 per 1,000 pages with no minimum commitment, a small pilot costs very little regardless. Check the current free allowance on Mistral's pricing page before planning around it.
It depends on the endpoint. For plain text, AWS Textract Detect Document Text is $1.50 per 1,000 pages against Mistral's $4, so Textract is cheaper. For invoices, Textract AnalyzeExpense costs $10 per 1,000 pages while Mistral Document AI returns structured fields for $5, so Mistral is roughly half the price.
Not from the OCR endpoint on its own. That returns the page as markdown text with layout preserved. To get typed fields such as vendor, invoice number, and total, you use the Annotations feature billed as Document AI, and you define the schema yourself. A purpose-built invoice tool returns those fields with no schema work.
Yes. Mistral OCR 4 ships as a single container for self-hosted deployment, which suits organizations with data residency or compliance requirements that prevent sending documents to a public API. Self-hosting shifts cost from per-page fees to infrastructure and operations.
If you have engineers, Mistral OCR at $2 to $5 per 1,000 pages plus your own parsing code is the lowest software cost. If you do not, a flat-rate invoice tool from $49 a month that already handles field extraction, line items, validation, and Excel export is cheaper overall, because it removes the developer time that dominates the total.
Comparing cloud OCR APIs on cost? See AWS Textract pricing per page.
See prebuilt invoice cost per page on Azure Document Intelligence.
Pricing the Google Document AI Invoice Parser? See the breakdown.
Prefer an API? Pull invoice data to JSON without per-page metering.
Capture full line-item tables, not just the totals.
Compare the top invoice extraction tools on price and accuracy.
Read PDFs, scans, and photos with AI-based invoice OCR.
Start turning your invoices into clean, structured spreadsheet data.
USD
per month
billed as
$288 yearly
Choose speed vs accuracy when extracting
| Base AI Faster | 2,500 pages |
| Pro AI Best accuracy | 500 pages |
Scale invoice extraction across your whole team with automation.
USD
per month
billed as
$888 yearly
Choose speed vs accuracy when extracting
| Base AI Faster | 10,000 pages |
| Pro AI Best accuracy | 2,000 pages |
Enterprise‑grade invoice extraction, security, and controls.
USD
per month
billed as
$ yearly
Choose speed vs accuracy when extracting
| Base AI Faster | pages |
| Pro AI Best accuracy | pages |
