Free invoice and receipt OCR APIs, compared.

Which invoice OCR APIs have a genuinely free tier in 2026?

Eight of the sixteen APIs on this page run a recurring monthly allowance rather than a trial: LlamaParse, Azure Document Intelligence, Tabscanner, Veryfi, Invoice Data Extraction, Parseur, Eagle Doc, and Google Document AI on one processor only. ParseRail is not among them, it runs no free tier. Three run trials and call it a free plan, Mindee for 14 days, DocuClipper for 14 days and up to 120 pages, Taggun for 30 days and up to 1,000 scans. Four more look recurring and are not. AWS Textract grants 100 pages a month but only for the first three months of a new AWS account; Nanonets grants $50 in credits once, after which a $100/month floor applies; Reducto is free to the first 15,000 credits and $0.015 a credit thereafter; and Lido grants 50 pages that never expire but never renew either. Lido carries a second catch that the free-tier column cannot show: its $29/month plan has no API at all, so a developer who spends the 50 free pages is looking at $7,000 a year, not $29 a month, to keep calling it from code.

The split matters because the two kinds answer different questions. A trial tells you whether the extraction is accurate enough on your documents, which is worth knowing and takes about an afternoon. A recurring allowance lets a low-traffic product sit in production without a bill, and lets an integration stay live while you decide about the rest of the year. If you are evaluating, any of the sixteen will do. If you are shipping something small, only the nine are relevant, and only six of those return validated invoice or receipt fields rather than page text.

How many invoices a month can you actually process for free?

Between 20 and 10,000 one-page documents a month, depending on the vendor and on how much of the work you want done for you. The eight recurring allowances that handle invoices, largest first:

1. LlamaParse10,000 pages a month. The largest free allowance here by a factor of twenty, and the one that does the least. LlamaParse parses, it returns document structure and text, not validated invoice fields. Basic mode spends one credit a page, so 10,000 credits is 10,000 pages; richer modes spend more. You still budget an extraction layer on top.
2. Azure Document Intelligence500 pages a month. The largest free tier that returns actual invoice fields, and it recurs indefinitely on the F0 tier. The catch is on the other side of it: Azure publishes every paid rate as a “$-” placeholder that defers to a signed-in calculator, so you cannot price your second month before you commit to it.
3. Tabscanner200 receipts a month. Receipts and nothing else. Tabscanner defines its unit plainly, “1 credit = 1 receipt scan”, and takes no card for the Starter plan. If receipts are your only document, this is the most generous no-card free tier that returns structured fields.
4. Veryfi100 documents a month. Labelled “Free Forever” on Veryfi’s own pricing page, and it covers any supported document type rather than one. The step off it is the steepest on this table: the next plan up is a $500/month minimum commitment.
5. ParseRailNone, a $20 pack is 250 invoices or 333 receipts. No free allowance at all as of September 2026, and nothing on this page should suggest otherwise. Credits are bought up front: an invoice spends 8, a receipt 6, a bank statement 12, at one cent a credit. The argument is entirely on the far side, $0.08 an invoice, published, no minimum, no plan, and billed only when extraction succeeds against your schema.
6. Invoice Data Extraction50 pages a month. No subscription and no monthly minimum on the paid side either, you buy credit packs when you need them, and it bills only on a successful process. The pack prices are not published.
7. Parseur20 pages a month. Parseur counts a whole email or spreadsheet as one page regardless of length, which makes 20 pages go further than it reads if your inputs arrive by email. Documents are retained 90 days on the free tier.
8. Eagle Doc20 pages a month. The same allowance as Parseur, and the cheapest paid step off any free tier here, €5.00 a month for 166 pages, excluding VAT. Prices are published in euro only.

Read that list in the right order and it stops being a ranking of generosity. The two largest allowances each cost you something the number does not show: LlamaParse gives you 10,000 pages of structure you then have to extract fields from, and Azure gives you 500 pages of real fields against a paid rate it will not publish until you sign in. The question a free tier should answer is not how long you can avoid paying. It is what the first paid document costs, and whether you can find that number out before you depend on it.

Does Google Document AI have a free tier?

Yes, but not on the parsers you would use for an invoice or a receipt. Google’s Document AI pricing page lists exactly one processor with a free band, the Enterprise Document OCR Processor, priced “0 count to 1,000 count $0.00 (Free)”, then $1.50 for the 1,000-to-5,000,000 count band and $0.60 above that, charged per 1,000 counts in the same way Google’s own worked example prices its Form Parser at “$30 / 1,000 pages”. Since Google defines one count as “up to 10 pages in a document”, that free band is worth up to 10,000 pages of OCR a month. What it returns is text and layout, not a supplier name, an invoice total, or a line-item array.

The parsers that do return those fields carry no free band at all. Google lists the Invoice parser at $0.10 / 1 count, the Expense parser (which its own page labels “formerly receipt parser”) at $0.10 / 1 count, and the Bank statement parser at $0.75 per classified document, flat rates from the first document, with no tiered $0.00 row above them. So the honest answer to “is Google Document AI free” is that a one-page invoice costs ten cents on your first call and every call after it, and the free allowance you may have read about elsewhere belongs to a different processor doing a different job. One term does work in your favour: Google states that if a document contains no eligible form, “processing doesn’t occur and no charge is made.”

What happens when the free tier runs out?

This is the number to shop on, and it is the one the free-tier comparisons leave out. Veryfi’s hundred free documents are followed by a $500/month minimum commitment that buys fewer than 5,000 documents, so the 101st document does not cost $0.16, it costs whatever $500 divided by your real monthly volume works out to. Mindee’s cheapest published plan is $29/month, or $312 for a year billed annually, and the smallest credit volume it sells is 500 a month for $30. Nanonets sets a $100/month floor the moment the $50 grant is gone. Azure publishes no rate to run out into at all.

Four vendors here let you cross the line without a step up. Eagle Doc’s first paid tier is €5.00 a month for 166 pages, excluding VAT, the cheapest paid step on this table. Invoice Data Extraction sells credit packs with no subscription and no minimum. AWS Textract charges $0.01 a page on AnalyzeExpense with no floor, which remains the cheapest headline rate in this market and we are not going to pretend otherwise. ParseRail charges $0.08 an invoice and $0.06 a receipt, published, no plan, no minimum, and billed only when extraction succeeds against your schema. There is no free allowance in front of that price, the entry point is a $20 pack.

Methodology

Every figure on this page was read from the vendor’s own public pricing page, never from a summary, a reseller, or a remembered number. Where a vendor publishes no figure, this page publishes none and says so. A free tier is counted as “recurring” only where the vendor’s own page states a monthly allowance; anything time-boxed is a trial here, whatever the vendor calls it.

Read fresh on 26 August 2026. Two vendors were added. Reducto (reducto.ai) lists its Standard plan as “Free” and “Up to your first 15K credits”, followed by “Then, $0.015 per credit after first 15K”. That is an allowance granted once, not a monthly one, so it is filed here as expiring rather than recurring. Its unit needs stating too: the page’s own FAQ says a call spends credits “based on the operation type, configurations, and the number of pages processed”, so 15,000 credits is not 15,000 pages and this page does not claim it is. Lido (lido.app) offers “50 free pages” with “No credit card required”, and its FAQ says “The free pages do not expire”, but they do not renew either, which is why it sits with the one-time grants rather than the recurring tiers. The fact that matters most for a developer is not on its pricing cards at all: API and workflow access are listed only against the Scale plan at “$7,000 / yr”, so the $29/month tier cannot be called from code. One Lido figure is contradicted on its own page and is therefore quoted nowhere here: the Enterprise card reads “From $35,000 / yr” while the FAQ below it reads “Enterprise plans start at $30,000/year”.

Read fresh on 13 August 2026. Veryfi publishes a “Free” plan at “Process up to 100 docs/mo” for “$0” marked “Free Forever”, with its Starter tier described as a “$500/mo minimum commitment”; separately its header offers a “Free 14-day trial, no credit card required”. LlamaParse lists a “Free $0 /month” plan that “Includes 10K credits”, with Starter at $50/month including 40K; its own FAQ describes “a generous provision of 10k free credits a month to all new users”. Parseur states “Start on our free tier, which includes 20 pages every month”, and its feature table gives the free tier a 90-day document retention period and one user. Tabscanner lists Starter at “$0 FREE” with “200 credits per month”, above the line “1 credit = 1 receipt scan. No credit card required for starter plan.” Google Document AI lists the Enterprise Document OCR Processor at “0 count to 1,000 count $0.00 (Free)” and its Invoice parser, Expense parser and Bank statement parser at $0.10/count, $0.10/count and $0.75 per classified document respectively, with no free band on any of the three, under the footnote “1 count equals up to 10 pages in a document”.

Carried from the register, with their own dates. Invoice Data Extraction (50 pages/month, no card, no subscription), Eagle Doc (20 pages/month at €0.00, €5.00/month Starter for 166 pages excluding VAT) and Nanonets ($50 one-time credits, $100/month floor) were read at source on 12 August 2026. Azure Document Intelligence’s 0-500 pages/month F0 tier was reconfirmed on 9 August 2026, it is the only Azure figure published as a number rather than a “$-” placeholder, which is why this page states Azure’s free tier and refuses to state its paid rate. AWS Textract (100 pages/month for the first three months, $0.01/page on AnalyzeExpense thereafter) and Taggun (30-day trial, up to 1,000 scans) were read on 7 August 2026. Mindee (no free tier, 14-day trial, $44/month billed annually as the published floor, 6,000 credits the smallest volume sold) and DocuClipper (no free tier, 14-day trial up to 120 pages) were read on 6 August 2026. All of these are held in this repository’s dated-claim register and re-checked against source on a daily cycle; a value that cannot be re-read is marked rather than quietly kept.

ParseRail’s own figures are the live platform rates: no free grant, one credit is one cent, credits are bought in packs from $20, and an invoice spends 8 credits, a receipt 6, a bank statement 12. This page carries no accuracy claims, how well extraction actually performs is scored in the open at /benchmarks. For what each API charges per document once you are paying, see invoice OCR API pricing compared; for receipts specifically, see receipt OCR APIs for developers; and for which API fits a given job rather than which is cheapest, see the best document extraction APIs in 2026; and for the monthly bill at a stated volume once a free allowance runs out, see how much an invoice OCR API costs per month.