Is Google Cloud Document AI free?

The Short Answer: Free Tier Exists, But Only for OCR

If you've been comparing document extraction tools and concluded Google Cloud Document AI has no free tier, you've been misled by outdated comparison posts. Google's official pricing page lists the Enterprise Document OCR Processor at $0.00 for the first 1,000 pages every month — a genuinely recurring free allowance, not a trial.

That said, "free" applies narrowly. The official pricing table shows that only OCR gets the free tier; every other processor — Form Parser, Custom Extractor, Layout Parser, and the specialized parsers — is billed from the very first page. Form Parser and Custom Extractor, for example, run $30 per 1,000 pages at entry volume.

Beyond the recurring tier, new customers get a one-time $300 credit that can be spent on Document AI or any other Google Cloud product, according to the product page. Combined with the free OCR allowance, that's enough to prototype a document pipeline at no real cost — as long as your use case stays within text extraction.

So why do so many competitor comparisons miss this? Several third-party sources state that Google Document AI has no free tier at all, while highlighting rivals' allowances. The likely explanation is that the free tier is OCR-only and easy to overlook, and Google's own product page doesn't mention it — only the $300 credit. The official pricing page should be treated as the authoritative source.

For context when shopping around, the free tiers look like this:

• Google Document AI: 1,000 free pages/month, but OCR only
• Azure Document Intelligence: 500 free pages/month across all features
• AWS Textract: 100 free pages/month for 3 months, new customers only

One caution before you budget around the free tier: once you move past OCR, costs climb quickly. Custom processors incur hosting charges of $0.05 per hour — roughly $438 per year for a single deployed processor version — and pretrained parsers bill in 10-page increments, so a one-page receipt costs the same as a ten-page one. Failed requests aren't billed, which offers some protection.

The practical takeaway: Document AI is free only if your job is plain text extraction at under 1,000 pages per month. For anything more — form parsing, custom field extraction, invoice processing — expect paid rates from page one, and model your costs by document page bands rather than raw page totals.

And if your workflow extends beyond extraction into routing, validation, or automating the post-OCR steps, agent skills can handle the rest of the pipeline. SkillzBench is building a free directory of verified agent skills plus a free MCP connector so you can browse, test, and wire those skills into your document workflows — worth exploring once the free OCR tier gets you started.

What Costs Money: Per-Processor Pricing and the 10-Page Billing Trap

The moment you step beyond OCR, the meter starts running — and Google's pricing structure contains a few surprises that catch even experienced cloud users off guard. Here's what actually costs money.

According to Google's official pricing page, Form Parser and Custom Extractor are priced at $30 per 1,000 pages for up to 1 million pages monthly, dropping to $20 above that threshold. Layout Parser runs $10 per 1,000 pages. There is no free allowance for any of these — paid pricing starts at page one.

Specialized parsers bill differently. Invoice, Expense, Utility, Driver License, and Passport parsers charge $0.10 per "count," where one count covers up to 10 pages. Bank statement parsing costs $0.75 per classified document, and Pay slip/W2 parsing costs $0.30.

The 10-Page Billing Trap

This is where budgets quietly break. Because specialized parsers bill in 10-page blocks, a one-page receipt costs exactly the same as a ten-page contract — $0.10 either way. Competitor analysis flags this as a key hidden cost, and Google's own examples confirm it: a 1–10 page document costs $0.10, while a 91–100 page document costs $1.00.

If your documents are short — receipts, IDs, single-page invoices — you're effectively paying a 10x premium per page. Model your costs by page bands, not raw page counts.

The Hidden Hosting Fee

Custom Extractor adds a charge most people miss: processor version hosting at $0.05 per hour. Google's own pricing documentation gives the worked example — one processor version deployed for a full year costs $438, and that's before a single page is processed. Deploy multiple versions for testing and production, and hosting costs multiply.

A few billing protections soften the blow:

• Failed requests (4xx or 5xx responses) are not billed.
• Documents with no eligible form are not charged — processing simply doesn't occur.
• Volume discounts kick in above 1M–5M pages, depending on the processor.

One competitor claims real total cost runs 3–5x the listed API price once you factor in Cloud Storage, Cloud Functions, and 40–80 hours of developer integration time — a vendor claim worth noting, though not independently verified.

For teams building full document pipelines, the extraction API is only step one — routing, validation, and post-processing automation often follow. Agent skills can automate the rest of this workflow, and SkillzBench is building a free directory and MCP connector to help you browse, verify, and test them.

How Google's Free Tier Compares to Azure and AWS

At first glance, Google's free tier looks like the most generous of the three major cloud document services — but the details tell a more nuanced story. Each provider structures its free allowance differently, and those differences matter depending on what you actually need to extract.

Google Cloud Document AI offers 1,000 free pages per month, but only for its Enterprise Document OCR Processor, which is listed at $0.00 on the official pricing page. Every other processor — Form Parser, Custom Extractor, Layout Parser, and the specialized parsers — is paid from the first page. So if you need structured field extraction rather than plain text recognition, Google's free tier doesn't apply.

Azure takes the opposite approach. Microsoft's Document Intelligence free tier (F0) includes 500 pages per month, but those pages can be used across all features, not just OCR, according to the Azure pricing page. Microsoft also offers new users a $200 credit and free services for 30 days.

AWS Textract is the most restrictive of the three. New AWS customers get 100 free pages per month for AnalyzeExpense, and only for three months, per a three-way comparison of the services.

Here's how the free allowances stack up:

• Google Document AI: 1,000 pages/month, OCR-only, no stated time limit
• Azure Document Intelligence: 500 pages/month, all features included
• AWS Textract: 100 pages/month, three months only, new customers
• Google also offers a one-time $300 credit for new Cloud customers

But free pages aren't the whole picture. An independent January 2025 benchmark by Businessware Technologies found Google's Invoice Parser ranked lowest in accuracy among the three: 82% field-level accuracy and just 40% line-item extraction accuracy, compared to Azure's 93% and 87%. The same benchmark put all three services at roughly comparable cost — around $10 per 1,000 pages for invoice processing.

That finding flips the value calculus. A free tier that misreads your invoice line items may cost more in manual correction time than a paid service that gets them right. Businessware's conclusion was blunt: Google's lower performance makes AWS Textract "a better value overall," despite similar pricing.

The practical takeaway: if your workload is pure text recognition at moderate volume, Google's 1,000-page OCR allowance is hard to beat. If you need field extraction, Azure's 500 all-feature pages deliver more usable capability per free page. And whatever you choose, remember that extraction is only half the workflow — routing, validation, and follow-up steps can be automated with agent skills, which SkillzBench is building a free directory and free MCP connector to help you browse, verify, and test.

Hidden Costs and How to Estimate Your Real Bill

The sticker price on Google's pricing page is rarely the number that lands on your invoice. Several costs hide behind the per-page rates, and teams that model only raw page counts routinely underestimate their bills.

The biggest hidden line item is integration. One competitor comparison claims real total cost runs 3–5x the listed API price once you add GCP project setup, Cloud Storage, Cloud Functions, and 40–80 hours of developer time — treat that as a third-party claim, not Google's own figure, but the underlying components are real. The same source cites Gartner research finding hidden costs in enterprise document processing contracts average 25–40% above list prices.

Then there are Google's own quietly accumulating charges. Custom processor hosting bills at $0.05/hour — Google's own example shows a single deployed version costs $438 over a year, and that runs whether or not you're processing documents. Cloud Storage for input files and output results adds a further, usually small, monthly cost.

Billing quirks matter as much as headline rates. Pretrained specialized parsers charge per "count," where one count covers up to 10 pages, so a one-page receipt costs the same as a ten-page contract. Google's own examples make this concrete: a 1–10 page document costs $0.10, while a 91–100 page document costs $1.00.

The good news: you are not billed for failed requests (4xx or 5xx responses), and documents with no eligible form are never charged, per Google's official pricing page. That's a meaningful protection when you're processing messy real-world inputs.

To estimate your real bill, model costs by page bands rather than raw page totals:

• Sort your documents into bands (1–10, 11–20, 21–100 pages) and apply the per-count rate for specialized parsers, or the per-1,000-page rate for OCR and Form Parser.
• Add $438/year for every custom processor version you keep deployed.
• Budget for Cloud Storage, Cloud Functions, and developer integration time — the components competitors flag as the true cost drivers.
• Strip out expected failed requests before multiplying by volume.

If your document pipeline extends beyond extraction — routing, validation, summarizing outputs — agent skills can automate the rest of this workflow. SkillzBench is building a free directory and MCP connector for browsing and testing exactly those skills, so the integration budget you estimate today can shrink as the ecosystem matures.

Getting Started Without Spending: Free Path and Automating the Rest

You don't need a budget to start extracting real value from Document AI — you need a plan for the free allowances. Google's official pricing table lists the Enterprise Document OCR Processor at "$0.00 (Free)" for the first 1,000 pages per month, and that allowance applies across standard, 1-year, and 3-year plan tiers (Google Cloud pricing). Plain OCR handles text extraction from PDFs, images, and scanned documents — supporting 200+ printed languages and handwriting in roughly 50.

Everything beyond OCR is paid from the first page, so use the one-time $300 new-customer credit to test the paid processors before committing (Google's product page). That covers Form Parser, Custom Extractor, and specialized parsers like the Invoice parser, which charges $0.10 per count — where one count equals up to 10 pages. Watch that billing block: a one-page receipt costs the same as a ten-page one, so model your costs by page bands, not raw page totals.

If you need structured field extraction from your own document formats, fine-tuning a Custom Extractor is possible from as few as 10 sample documents. But remember the hosting trap: custom processor versions bill at $0.05/hour, which Google's own example puts at roughly $438 per year for a single deployed version. Undeploy when you're done testing.

Know the operational limits before you build:

• Synchronous requests don't support documents over 10 pages
• Batch requests handle documents up to 200 pages each
• You aren't billed for failed requests (4xx or 5xx errors)
• No charge applies when an input document contains no eligible form

Extraction is only half the workflow, though. Once text and fields come out of Document AI, someone still has to route documents to the right team, validate the extracted values, and push results into your systems of record. That's where agent skills come in — instruction packages that let AI coding agents and chat-app agents handle these post-extraction steps automatically. The catch is quality: published research found roughly 26% of published agent skills carry at least one vulnerability, so picking them blindly is risky.

SkillzBench is building a free, independent directory of agent skills with a free MCP connector that links any agent to the skills you choose — every skill version-pinned, security-scanned, and benchmarked with its Lift published in percentage points. Everything on the platform will be free, funded by blog advertising rather than user fees. If you're assembling a document pipeline on Document AI's free tier, it's the place to find vetted skills for routing and validation without adding spend.