Pay per page. No subscription. No monthly minimums. No expiring credits. You pay only for pages that were successfully analyzed — failed documents don't cost anything.
Claude Opus 4.7 for extraction, followed by a second Opus pass that audits every field across the entire document. The audit explicitly checks:
Flags use hedged language ("appears inconsistent", "may indicate", "warrants human verification") and quote the conflicting numbers verbatim so a reviewer can spot-check in seconds. Good for:
Claude Sonnet 4.6 single-pass extraction, no audit. Citations still available as an opt-in. Good for:
See the Quick vs. Verified section of the FAQ for more on what the audit pass catches, or the citations section for how source quotes work.
On any run, you can opt into source citations. Every extracted value comes back with the page number and the exact verbatim quote it was drawn from. The CSV gets a (page) and (source quote) column next to each field, and on the results page each cell shows a p.N chip that opens a popover with the quote and a copy-to-clipboard button. On the Verified tier, the audit pass additionally validates each citation against its value — catching cases where the model fabricates an answer that isn't actually supported by the quoted text. Citations add no per-page charge.
Two-tier minimum designed so you can cheaply iterate on a sample before committing to a big run:
For typical contract batches, neither minimum is ever reached — you just pay the per-page rate.
After a run completes, the results page shows a Tune & Re-run button. Change fields, custom prompt, or tier and run again against the same uploaded documents — no re-upload. Each tune run shows up as its own tab on the results page so you can compare runs side by side. A Compare button renders a merged table with diffs highlighted, making prompt tuning on a 2-doc sample a $0.50/run cycle instead of an expensive full-batch gamble.
| Batch | Verified ($0.15/page) | Quick ($0.10/page) |
|---|---|---|
| 1 × 3-page contract (prompt-tuning sample) | $0.50 (small-batch min) | $0.50 (small-batch min) |
| 1 × 15-page contract | $2.25 | $2.00 (minimum) |
| 1 × 50-page commercial agreement | $7.50 | $5.00 |
| 10 × 20-page vendor contracts | $30.00 | $20.00 |
| 50 × 15-page SaaS agreements | $112.50 | $75.00 |
| 100 × 20-page contracts (2,000 pages) | $300.00 | $200.00 |
| 200 NDAs × 3 pages (600 pages) | $90.00 | $60.00 |
| 100 commercial leases × 40 pages (4,000 pages) | $600.00 | $400.00 |
| 500 × 30-page contracts (15,000 pages) | $2,250.00 | $1,500.00 |
PDFs: counted by actual page count. A 27-page PDF costs 27 pages regardless of how much content is on each page.
Word, text, HTML, CSV, XML, JSON, Markdown: estimated at roughly 3,000 characters per page. A short DOCX counts as 1 page; a long one is counted proportionally.
Images (PNG, JPG, GIF, WebP): counted as 1 page each. A scanned contract sent as 10 separate image files counts as 10 pages.
ZIP archives: the archive itself isn't charged; each file inside the ZIP is counted by its own type (PDF by page, DOCX by character count, etc.). ZIPs of ZIPs aren't recursively unpacked.
Failed documents: if processing fails (corrupt PDF, password-protected, heavily degraded scan), that document isn't charged.
Most document parsing tools in this market use monthly subscriptions with renewable credits. That model works when you have steady, predictable volume — you'll fully use your credits each month, and the per-page rate comes out competitive.
It works poorly when volume is uneven or project-based. Process 250 contracts one weekend, then nothing for two months, and you've paid for three months of subscription credits that evaporated unused.
ContractParser uses pay-per-page specifically to fit project-based and one-time work:
For truly steady monthly volume, a subscription-based competitor may be cheaper per page at scale. See our Airparser comparison or Docparser comparison for detail on where each model wins.
All payments are processed by Stripe. We never see or store your card number, CVV, or any other card data.
At checkout, your card is authorized (funds placed on hold) for the expected cost based on the page count of your batch. This is a hold, not a charge — it doesn't affect your balance at that point.
After processing completes, we capture only the actual cost of documents that were successfully analyzed. Documents that failed to process (corrupt PDF, password-protected, heavily degraded scan) aren't charged.
If every document in a batch fails to analyze, the authorization is released and you pay nothing. The remainder of an authorization above the captured amount is released back to your card automatically.
Receipts are delivered by Stripe to the email you provide at checkout. For invoicing with your company name and billing address, Stripe's hosted invoice is sufficient for most procurement systems.
The $0.10/$0.15 pricing applies at all volumes through the web interface. For enterprise customers with specific requirements — volume discounts, custom retention windows, audit logging, DPAs, vendor security questionnaires, named AI provider under NDA, or SSO — contact us at support@grovestreams.com.
For developers who need API access for programmatic extraction or integration, the API is on the roadmap. Email us to be notified when it's available.