Is Google Document AI free?

The Real Answer: Free for 1,000 Pages of OCR, Paid After That

Here is the short version: Google Document AI is free for exactly one thing — the first 1,000 pages per month of Enterprise Document OCR. Everything else costs money from your very first request. That's the answer straight from Google's official pricing page, and it's the detail most comparison articles get wrong or leave out entirely.

The free tier applies only to the Enterprise Document OCR Processor, and it holds across all consumption models. So if your need is basic text extraction — turning scanned PDFs and images into machine-readable text — you can process up to 1,000 pages a month at $0.00. After that, OCR runs $1.50 per 1,000 pages for the first 5 million pages, dropping to $0.60 per 1,000 pages beyond 5 million.

Every other processor is paid from the first request. That means if you want structured extraction rather than plain OCR, there is no free tier at all:

• Form Parser and Custom Extractor: $30 per 1,000 pages (up to 1 million pages)
• Layout Parser: $10 per 1,000 pages
• Specialized parsers (Invoice, Expense, US Driver License, Passport): $0.10 per document, where one document counts as up to 10 pages
• Bank Statement Parser: $0.75 per document
• Custom Splitter/Classifier: $5 per 1,000 pages

One quirk worth knowing: Google's pricing page doesn't explicitly state whether the 1,000-page free allowance resets monthly, so treat the reset period as implied rather than guaranteed.

There's also a hard prerequisite that surprises many first-time users: you must enable billing on a Google Cloud project to use Document AI at all — even for the free OCR tier. Google's own tutorial documentation is unambiguous on this point, and the OCR setup guide lists verifying billing as a prerequisite step. In other words, "free" here still means attaching a payment method.

The practical workaround for evaluating paid processors is Google Cloud's $300 free trial credit for new customers, which you can spend on Form Parser, Invoice Parser, or anything else to test before committing. Just note that this is a general Google Cloud offer, not a Document AI-specific free tier.

If your plan goes beyond raw OCR — routing extracted data, validating fields, triggering downstream actions — agent skills can automate the rest of that workflow. SkillzBench, currently in pre-launch, is building a free directory and MCP connector for security-scanned, benchmarked agent skills, so you can browse and test options before wiring anything into production.

What You'll Actually Pay: Document AI Pricing Breakdown

Once the free tier runs out, Document AI pricing is refreshingly transparent — but the numbers vary dramatically depending on which processor you use. Here's what the official figures actually look like.

According to Google's pricing page, Enterprise Document OCR costs $1.50 per 1,000 pages for the first 5 million pages, dropping to $0.60 per 1,000 pages beyond that mark. Optional savings plans shave 10% off with a 1-year commitment and 20% with a 3-year commitment.

The more powerful parsers cost considerably more. Form Parser and Custom Extractor both run $30 per 1,000 pages up to 1 million pages, then fall to $20 above that threshold. Google's own examples put this in perspective: 100 pages through Form Parser costs $3/month, while 6 million pages would run $130,000/month.

Specialized parsers bill per document instead, where one document counts as up to 10 pages:

• Invoice, Expense, Utility, US Driver License, and US Passport parsers: $0.10 per document
• Pay Slip and W2 Parser: $0.30 per document
• Bank Statement Parser: $0.75 per document
• Procurement and Lending Splitter/Classifier: $0.05 per classified document

Page-count rules matter for planning. Synchronous requests don't support documents over 10 pages, while batch processing handles up to 200 pages per document. File types convert predictably: a DOCX or HTML file up to 3,000 characters counts as one page, and each XLSX tab or PPTX slide counts individually.

Two line items surprise first-time users. Custom processor hosting runs $0.05/hour — roughly $438 per year for every deployed version — and a provisioned service tier costs about $0.41/hour. Google confirms on its pricing page that failed requests (4xx or 5xx response codes) aren't billed, and documents classified as "other" by the Procurement Splitter & Classifier are also free.

Be aware that total costs can stretch beyond API fees. Document AI is designed to work alongside other Google Cloud products like Cloud Vision and the Natural Language API, which carry separate charges — and a competitor analysis argues real-world costs often exceed the listed price once engineering time is factored in, though that claim carries obvious promotional bias.

If your goal is automating the full workflow around these documents — routing extracted data, validating fields, triggering downstream actions — agent skills can handle much of it. SkillzBench is building a free directory and MCP connector for security-scanned, benchmarked skills, so you can test which ones genuinely improve your pipeline before committing.

The $300 Trial Credits and Hidden Costs to Budget For

For anyone evaluating paid processors like the Invoice Parser or Form Parser, the practical "free" path isn't the free tier at all — it's Google Cloud's $300 in trial credits for new customers, described by Google as having "no automatic charges, no commitment." That's enough headroom to test real documents against real processors before a single dollar leaves your budget.

The trial is a general Google Cloud offer rather than a Document AI-specific perk, so it covers the full stack you'd need to experiment. One caveat: Google's own tutorials require you to enable billing on your Cloud project to use the service at all, even while spending credits. And once the trial ends, the meter keeps running on whatever you've deployed.

That's where total cost of ownership comes in. The per-page API price is only the visible line item. Google's pricing page notes that Document AI "is intended to be used with other Google Cloud products" — Cloud Vision, Cloud Natural Language, the Gemini platform — each with its own billing. There's also processor hosting to consider: custom processor versions cost $0.05 per hour, which works out to roughly $438 per year for a single deployed version, whether or not you're processing documents. Google's own codelab warns users to shut down notebook instances or delete projects entirely "to avoid incurring charges."

Budget for these hidden line items:

• Companion services like Cloud Vision and Natural Language, billed separately when Document AI feeds into a larger pipeline
• Custom processor hosting at ~$0.05/hour per deployed version
• Provisioned service tiers at ~$0.41/hour for guaranteed capacity
• Engineering time for setup, storage, and pipeline glue code

On that last point, one competitor comparison site claims the real total cost is "typically 3–5x the listed API price" once GCP infrastructure and 40–80 hours of engineering are included. Treat that as opinion with promotional stakes, not verified fact — but the directional warning is fair. Google's own pricing examples show how quickly scale compounds: 100 pages through Form Parser costs $3 per month, while 6 million pages costs $130,000 per month.

It's also worth seeing Google's free offering in context. AWS Textract gives new customers 100 free pages per month for just three months, while Azure Document Intelligence offers a 500-page monthly free tier. Google's 1,000-page OCR allowance is more generous, but it only covers one processor type.

If your ambitions extend beyond extraction — routing invoices, validating fields, triggering workflows from the extracted data — that downstream automation is where agent skills come in. SkillzBench is building a free directory of benchmarked, security-scanned skills plus a free MCP connector, so you can browse and test options without adding another line item to the budget.

How to Evaluate Document AI for Free — and Automate the Rest

Testing Google Document AI without spending a dime is possible, but it requires careful setup — the service cannot run at all unless billing is enabled on your Google Cloud project, even when you're only using free pages. Google's own tutorials are blunt about this: "You must enable billing in Cloud Console in order to use Google Cloud resources."

Here's a practical path to evaluate it at zero cost:

• Create a Google Cloud project and enable billing — this is a hard prerequisite for using Document AI.
• Claim the $300 in free trial credits available to new Google Cloud customers, described as "no automatic charges, no commitment."
• Test your highest-volume documents against the Enterprise Document OCR Processor first, since its first 1,000 pages per month are free.
• Use paid processors like Form Parser or Invoice Parser sparingly, charged against your trial credits.
• Shut down any notebook instances or delete the project when finished — Google's codelab explicitly warns this is needed "to avoid incurring charges."

Once your free OCR allowance is exhausted, costs scale quickly. Beyond the 1,000-page threshold, standard OCR pricing is $1.50 per 1,000 pages for the first 5 million pages, while paid processors cost far more: $30 per 1,000 pages for Form Parser or Custom Extractor, and $0.10 per document for specialized parsers like the Invoice Parser. Google's own pricing example shows that processing just 100 pages monthly through Form Parser runs $3 per month — and 6 million pages would cost $130,000 monthly.

Model your costs by page bands rather than raw monthly totals. One comparison of cloud document extraction services recommends this approach because invoice economics depend on whether documents stay within the first 10-page block, and Google's synchronous requests don't support documents over 10 pages at all — batch processing handles up to 200 pages. Also budget for extras beyond the API itself: custom processor hosting runs $0.05 per hour (roughly $438 per year per deployed version), and Document AI is designed to work alongside other Google Cloud products that bill separately. One competitor analysis claims total real-world costs can reach 3–5x the listed API price once setup, storage, and engineering time are included — treat that multiplier as a vendor's opinion, but the underlying warning about hidden overhead is worth heeding.

The good news: failed requests (4xx or 5xx) are not billed, so botched test runs won't drain your credits.

Once your documents are extracted, the remaining work — routing invoices to the right team, validating extracted fields, pushing data into downstream systems — is where agent skills come in. SkillzBench is building a free directory of AI agent skills and a free MCP connector that lets you attach vetted skills to any agent, with security scanning and benchmarking built in. That matters because published research found roughly 26% of agent skills carry at least one vulnerability. If you want to automate the rest of your document workflow once extraction is done, browsing and testing skills on SkillzBench will be free — the platform is in pre-launch, with its waitlist open now.