Datalab Pricing: Marker OCR API Cost per Page

Datalab pricing is per processor, per 1,000 pages, and every plan pays the same rate card. Converting a PDF to markdown with Marker costs $4 per 1,000 pages in fast or balanced mode and $10 in accurate mode. Pulling structured fields out costs $6 per 1,000 pages on the turbo and fast tiers, $15 on balanced and $20 on accurate. The Free plan includes a $20 monthly allowance on a work email ($10 on a personal one), the Team plan is $400 a month with $400 of usage included, and Enterprise is quoted.

Those figures come from the Datalab pricing page and its FAQ. What follows turns them into what a buyer needs before signing up: the full rate card in dollars, the surcharges that are easy to miss, what a month of receipts or invoices costs, and how Datalab sits against Mistral, Reducto, LlamaParse and Extend on the same job.

How much does Datalab cost?

Datalab sells three plans. None of them changes the per page rates. What you buy with Team is throughput, compliance paperwork and support.

The allowance is a dollar figure, not a page count. Usage draws it down first at the rates below, anything past it bills at the same rates, and it resets each cycle without rolling over. You do not need a card to use the Free allowance; you add one to keep going past it. Subscriptions are month to month and cancel from the dashboard.

One correction to what you will read elsewhere: several third party write-ups still say a new Datalab account comes with $5 in free credits. The pricing page now lists a recurring $20 or $10 monthly allowance instead, which is a materially better deal for anyone evaluating the API on real documents.

What does each Datalab processor cost per 1,000 pages?

This is the full rate card as published, in US dollars per 1,000 pages. The Marker name applies to the Convert processor, which is what most people mean by the Datalab OCR API.

Add-ons stack on top of a run: chart understanding is $3 per 1,000 pages, infographic parsing $4, word level bounding boxes with confidence scores $3, and the bundle of word, table cell and list bounding boxes $9. Track changes on a Word document is $6 per 1,000 pages. Cross page merging, which stitches a table that breaks across a page boundary back together, is in beta and bills only a variable compute charge described as typically a few cents per document.

What costs extra on top of the Datalab rate?

Five rules change the bill, and none of them appear in the headline rate.

• Extraction fees on balanced and accurate. The vendor says around 5 percent of balanced and accurate extractions bill above the fixed rate, when the extraction agent has to work through a long document or a schema that pulls many repeated rows. When it applies it is typically a dollar or two per document, more for a long or dense one. Turbo and fast extraction carry no such fee. Schemas over 750 input fields are not supported at all.
• Spreadsheets bill by cell, not page. On the simple path 2,500 cells count as a page, capped at $0.60 per sheet. On the intelligent path 500 cells count as a page, and the FAQ confirms that 500 cells is one page for rate purposes. Datalab picks the path automatically.
• EU processing adds 25 percent to the summed total on Free and Team. US processing is the base price.
• Two discounts, never both. Opting into data retention takes 25 percent off. The startup program takes 33 percent off both the per page rates and the Team fee for twelve months, for companies under three years old with less than $15 million raised and under $5 million in annual recurring revenue, where document processing is core to the product. You choose one.
• Custom processors past the allowance cost $5 each on Team after the four included each month.

There is also a migration offer of up to $5,000 in credits for moving a workload from another provider, and the underlying models are open source and free for research, personal projects and startups under $2 million in funding or revenue, which matters if you are weighing self hosting against the hosted API.

Is Datalab free?

Partly. The Free plan gives you $20 of usage a month on a work email, or $10 on a personal one, against the full API with every processor at the same rate card. At $4 per 1,000 pages that is 5,000 pages of Marker conversion a month for nothing, or about 3,300 pages of turbo extraction. No card is required until you go past the allowance. There is no time limit on the Free plan, only the 25 request per minute rate limit and community support.

What does Datalab cost for receipts and invoices?

Here is the arithmetic for a finance workload, assuming one page per receipt and US processing. Convert alone gives you markdown, which is not the same as a vendor, date, total and tax field, so the realistic receipt path is Extraction.

Two things sit outside those numbers. The $20 Free allowance covers 500 receipts a month on every tier before fees, and at 2,500 receipts on turbo you are still under it. The second is the Team fee: at $400 a month you are paying for 400 requests a minute, a BAA and Slack support, not for cheaper pages, so a finance team with a few thousand receipts a month has no reason to move up from Free unless the rate limit bites.

The cost that does not appear on the page is the schema. Extraction returns the fields you define, which means someone on your team writes and maintains the receipt schema, handles multi page invoices, decides what to do with the 5 percent that bill extra, and builds the export to your accounting system. That is a developer job. If you have one, Datalab is among the cheapest ways to get structured receipt data at scale. If you do not, the per page price is not the price you will pay.

How does Datalab pricing compare with Mistral, Reducto, LlamaParse and Extend?

All of these publish a per page or per credit rate. The table restates each vendor in dollars per 1,000 pages for the two jobs that matter for receipts: raw conversion and field extraction. Vendor rates are from each pricing page; where a vendor prices in credits the conversion is ours.

On raw conversion Datalab ties Mistral at $4 and is only beaten by LlamaParse fast, which returns spatial text without markdown. On extraction only Mistral Document AI at $5 lists a lower rate than Datalab turbo at $6, and even Datalab balanced at $15 sits under Reducto and Extend performance. The catch is the plus fees on balanced and accurate, which none of the others apply in the same way. The per vendor detail is on our Mistral OCR pricing page, the Reducto pricing breakdown, the LlamaParse pricing page and the Extend AI pricing article, and the whole field is lined up on the OCR API pricing comparison.

Who should pay for Datalab, and who should not

Datalab is built for engineering teams that need documents turned into model ready text or structured JSON at volume: RAG pipelines, training data, agent workflows, compliance archives. The open source models, the VPC and air gapped options, the per field citations and the bounding box add-ons are the reasons to choose it, and every one of them assumes a developer on the other end. If document processing is core to what you are building, the startup discount and the recurring free allowance make it one of the cheapest APIs to start on. If you need the pipeline built but do not have the engineer, the fastest route is usually to hire an AI developer for the integration rather than to learn the SDK yourself.

If you are a bookkeeper, an accounts payable clerk or a small finance team, the arithmetic changes. Datalab charges $6 per 1,000 pages for turbo extraction, but that buys JSON against a schema you wrote, with no line item table, no Excel, no QuickBooks file and no review screen. Our Starter plan is $49 a month for 2,500 pages ($19.60 per 1,000), which is more per page and less per finished spreadsheet, because the schema, the line items, the sales tax split and the export are already done. The bulk receipt scanner takes up to 50 files in one upload, and teams that want the API without writing a schema can look at the receipt OCR API. Which model family is doing the reading, and why it matters for accuracy, is covered on our LLM OCR page.

Frequently asked questions

How much does Datalab cost per page?

Converting a page with Marker costs $0.004 in fast or balanced mode and $0.01 in accurate mode. Extracting structured fields costs $0.006 per page on turbo and fast, $0.015 on balanced and $0.02 on accurate, with a possible surcharge on the last two for long or dense documents. Rates are identical on the Free, Team and Enterprise plans.

What is included in the Datalab Team plan for $400?

Team includes $400 of usage a month against the same rate card, then meters anything beyond it. The fee buys 400 requests a minute and 400 concurrent requests instead of 25, four custom processor creations a month, a BAA and DPA, admin MFA enforcement and email plus Slack support. It does not lower the per page rates.

Does Datalab charge for pages that fail?

The FAQ says a failed form fill bills nothing, and cached results on cross page merging bill nothing extra. For the other processors the page count in each request breakdown is what you are billed on, so check the breakdown on your first few runs rather than assuming.

Is the Datalab API cheaper than running Marker yourself?

The models are open source and free for research, personal use and startups under $2 million in funding or revenue, so self hosting costs you GPU time and engineering hours rather than a license. Larger companies need a commercial arrangement. At $4 per 1,000 pages the hosted API is cheap enough that most teams only self host for data residency or air gap reasons, which Enterprise also covers.

Does Datalab read receipts and invoices?

Yes, through the Extraction processor with a schema you define, or through a custom or agent processor with natural language rules. It returns JSON, not a spreadsheet or an accounting file. If the goal is receipts in Excel or QuickBooks without writing code, a receipt specific tool with the schema and export already built is the shorter path.