How Much Do Document Extraction APIs Cost Per Page? 2026
The short answer
Plain OCR costs $0.0006 to $0.004 per page; extracting structured fields into JSON costs $0.005 to about $0.06 per page. The roughly 100x spread is the difference between reading text off a page and understanding it. As of October 10, 2026, the per-page list prices are:
| API | What you get | Per page | Per 1,000 pages | Volume / batch discount |
|---|---|---|---|---|
| AWS Textract — Detect Document Text | Text and lines | $0.0015 | $1.50 | $0.0006 after 1M pages/month |
| Google Document AI — Enterprise Document OCR | Text, layout | $0.0015 | $1.50 | $0.0006 above 5M pages/month |
| Azure Document Intelligence — Read | Text | $0.0015 | $1.50 | Commitment tiers; 500 pages/month free |
| LlamaParse — Fast tier (1 credit) | Spatial text | ~$0.00125 | ~$1.25 | 10,000 free credits/month |
| LlamaParse — Cost-effective / Agentic / Agentic Plus | Markdown, tables, charts | ~$0.004 / $0.0125 / $0.056 | ~$3.75 / $12.50 / $56 | Plan credits |
| Mistral OCR 4.1 | Markdown, tables, bounding boxes | $0.004 | $4 | Batch $2 |
| Mistral Document AI (annotated pages) | Schema JSON | $0.005 | $5 | — |
| Google Document AI — Layout Parser | Layout, chunks | $0.01 | $10 | — |
| Azure — Layout / prebuilt (invoice, receipt, ID) | Fields | $0.01 | $10 | — |
| Reducto — r-1 Parse | Layout, tables, markdown | $0.01 | $10 | Batch −20%; $150 free |
| LlamaParse — Extract (Cost-effective / Agentic) | Schema JSON, citations | ~$0.01 / $0.031 | ~$10 / $31 | Plan credits |
| Textract — Tables / Queries | Tables or Q&A fields | $0.015 | $15 | $0.01 tables after 1M |
| Reducto — Extract | Schema JSON, parsing included | $0.02 | $20 | Batch −20% |
| Azure — custom extraction | Trained fields | $0.03 | $30 | — |
| Reducto — Deep Extract | Agentic extraction | $0.04 | $40 | — |
| Textract — Forms | Key-value pairs | $0.05 | $50 | $0.04 after 1M |
| Textract — Tables + Forms + Queries | All three | $0.07 | $70 | $0.055 after 1M |
LlamaParse prices are converted from credits at $1.25 per 1,000 credits: Fast 1, Cost-effective 3, Agentic 10 and Agentic Plus 45 credits per page for parsing; Extract defaults to 8 credits (Cost-effective) or 25 (Agentic) per page including the parse.
What a real workload costs
100,000 pages a month of scanned reports, text only: Textract, Google or Azure Read $150; LlamaParse Fast about $125; Mistral OCR 4.1 $400 ($200 in batch).
10,000 invoices a month into JSON: Mistral Document AI $50; Azure prebuilt invoice $100; Reducto Extract $200 (minus 20% in batch, and the first $150 free); Textract Forms + Queries about $650.
Two million pages a month of plain OCR: Textract $2,100 ($1,500 for the first million at $0.0015 plus $600 for the next at $0.0006), which is AWS’s own published example.
Why the expensive tiers can be cheaper
Price per page is not price per usable field. A $0.0015 OCR call on a rotated, multi-column scan returns text in the wrong order, and someone pays to fix it. Tools priced at $0.01–$0.04 per page run layout models and an LLM pass, return citations and confidence scores, and route only low-confidence fields to review. Benchmark on 200 of your own documents and compare cost per correct field, not list price.
Three cost traps:
- Add-ons stack. Textract charges Tables, Forms and Queries separately; Azure charges extra for high-resolution, formula and query-field add-ons; LlamaParse adds 3 credits per page for layout extraction.
- Per-document vs per-page. Google’s prebuilt invoice, receipt and ID parsers bill per document of up to 10 pages, so short documents cost more per page than the table suggests.
- Re-parsing. LlamaParse caches results (a re-parse costs 0 credits); most others bill every call. Store the output.
Which to choose by budget
- Lowest cost at volume: Textract or Google Enterprise OCR at $0.0006 per page past the first tier.
- Best value for RAG ingestion: Mistral OCR 4.1 at $2 per 1,000 pages in batch, or LlamaParse Cost-effective.
- Invoices and IDs inside Azure or Google: the prebuilt models at $0.01 per page.
- Hard documents where errors are expensive: Reducto Extract or Deep Extract.
- Zero API cost: Docling or PaddleOCR on your own GPU.
For accuracy rankings rather than price, see our best OCR and document extraction APIs and best enterprise OCR vendors. LLM costs for the post-OCR step are in our current API prices.
Last verified: October 10, 2026. Google Document AI and Azure list prices were confirmed through their pricing pages and documentation; check your region before committing.