Hosted PaddleOCR-VL from $1 per 1,000 pages
PaddleOCR-VL-1.6 is an open-weight hosted lane on the OpenParser OCR router at $1 per 1,000 pages. New accounts get $10 free credits. Switch to Mistral, Azure, Google, or Textract on the same API when you need them.
Parsing and extraction pricing
Parse and extract endpoints meter the same way on every marketing surface: pages for parse, tokens for extract.
/parserequestcurl https://api.openparser.dev/parse \
-H "Authorization: Bearer op_live_…" \
-H "Idempotency-Key: $(uuidgen)" \
-F 'request={"ocr_model":"paddleocr-vl-1.6","output_format":"openparser@1"};type=application/json' \
-F "file=@statement.pdf"{
"output_format": "openparser@1",
"document_id": "doc_3f9a2c",
"page_count": 1,
"markdown": "## Payment Authorization\n| Item | Amount |\n| --- | --- |\n| Total | $4,318.20 |",
"blocks": [
{
"index": 0,
"page_number": 1,
"kind": "text",
"text": "## Payment Authorization"
},
{
"index": 1,
"page_number": 1,
"kind": "table",
"table_html": "<table><tr><th>Item</th><th>Amount</th></tr><tr><td>Total</td><td>$4,318.20</td></tr></table>",
"bbox": {
"left": 14,
"top": 53,
"right": 146,
"bottom": 123
},
"confidence": 0.98
}
],
"regions": [],
"contents": [],
"chunks": []
}Model volume in the console.
Create keys, watch page usage, and test documents in Open Studio before you commit to monthly volume.
{ total: "$4,318.20" } b4Layout OCR list prices, not text-only SKUs.
This page compares public list rates for layout-and-tables OCR. Output shape still decides whether a cheap SKU fits the job.
Layout OCR from $1 per 1,000 pages
OpenParser lists at $1 per 1,000 pages on open-weight hosted models in the layout-and-tables lane, the lowest rate among layout rows on this page. New accounts get $10 free credits.
OCR confidence on each block
Parsed blocks can include OCR confidence. Route uncertain pages to review before they hit your bill or your index.
Hosted or self-hosted PaddleOCR-VL
PaddleOCR-VL is open weights. Hosted OpenParser is the API path; self-hosting replaces page fees with GPU and ops cost.
Extraction is a second meter
Parse pricing stops at blocks and markdown. Schema fill adds the token cost of whichever llm_model you choose.
Citations stay cheap to audit
Extracted values cite block indexes from the parse, so review cost stays tied to real regions, not invented boxes.
Citation validation
An id that is not in the parse fails to resolve. You catch mismatches before they silently bill bad data downstream.
$10 free credits, then $1 per 1,000.
Parsing is a flat page rate on open-weight hosted models. Extraction adds the token cost of whichever model you select.
- No minimums
- Sync, async, and batch
- Typed blocks + markdown
- Block citations on extract
Put your page volume next to list rates.
Switch between basic-text and layout lanes. OpenParser applies $10 in free credits at signup (about 10,000 pages at $1/1,000); competitors use published per-1,000 rates.
Parse cost for your page volume
Structure-preserving parse. OpenParser’s layout output lives here.
100,000 pages · first 10,000 pages from $10 credits · then $1 / 1,000
- OpenParserincludes ~10,000 pages from $10 free credits$90
- Mistral OCR 4$4 / 1,000$400
- Azure Layout$10 / 1,000$1,000
- Google Layout Parser$10 / 1,000$1,000
- AWS Textract Tables + Layout$15 / 1,000$1,500
- Reducto (standard parse)$15 / 1,000$1,500
- · Parse cost only. OpenParser extraction token charges are excluded.
- · Competitor rows use the synchronous public list rate shown in the table.
- · Volume tiers (for example AWS / Azure / Google after 1M pages) are not modeled.
- · The calculator does not apply batch discounts such as Mistral's published 50% batch rate.
- · OpenParser applies $10 in free credits once to the entered volume (about 10,000 pages at the cheapest open-weight hosted rate).
- · One page = one billed page as each vendor defines it. Scanned PDFs and native digital pages may bill differently across vendors.
Two OCR jobs, two billing
lanes.
Public synchronous list prices checked 2026-07-24, with published credit rates converted to a per-page equivalent where needed. Output notes describe the published task, not a quality ranking.
Basic text / partial structure
Text and line boxes, or partial table markup. Cheaper hyperscaler SKUs live here.
| parser | list / 1k | output |
|---|---|---|
| Azure ReadS0 Read, first 1M pages / month | $1.50 | text and line boxes only |
| Google Enterprise Document OCRFirst 5M pages / month | $1.50 | text and line boxes only |
| AWS Textract Detect Document TextDetect Document Text, first 1M pages / month | $1.50 | text and line boxes only |
| Mistral OCR 3Synchronous API list; batch is 50% off per Mistral | $2 | Markdown/HTML tables, no typed blocks |
Layout + tables
Structure-preserving parse with layout and tables. OpenParser hosted PaddleOCR-VL output sits in this lane at the lowest list rate among these rows.
| parser | list / 1k | output |
|---|---|---|
| OpenParser (PaddleOCR-VL-1.6)$10 free credits; then $1 / 1,000 parsed pages on open-weight hosted | $1 | layout, tables, and figures |
| Mistral OCR 4Synchronous API list; batch is 50% off per Mistral | $4 | layout, boxes, and tables; no figures |
| Azure LayoutPrebuilt Layout / Document models | $10 | layout and tables; no figures |
| Google Layout ParserIncludes initial chunking | $10 | layout and tables; no figures preserved |
| AWS Textract Tables + LayoutAnalyze Document Tables; Layout free with Tables | $15 | layout and tables; no figures |
| Reducto (standard parse)$0.015 / credit after free allotment; 1 credit / standard page | $15 | full layout (closest equivalent) |
Match the lane to the job.
List price alone does not decide the stack. Output shape and billing mode still move the choice.
Different SKUs, different jobs
AWS Detect, Azure Read, and Google Enterprise OCR return text and line boxes. Layout and Tables SKUs add structure and cost more. Mixing those rates under one “cheapest OCR API” label hides the gap.
Synchronous list vs batch
This page uses synchronous public list rates. Some vendors publish batch discounts (for example Mistral’s 50% batch rate). Applying those discounts can reorder the table for high-volume jobs.
Credit meters vs page meters
Reducto bills in credits that scale with complexity. The $/1,000 figure converts a stated credit rate for a standard page. Complex pages cost more on that platform.
Extraction is a second bill
OpenParser parse pricing stops at blocks and markdown. Filling a JSON schema adds model token cost. Other vendors may fold schema extract into a higher page rate or a separate product.
Prices are dated
Every figure was re-checked against the linked vendor pages on 2026-07-24. Re-check those sources before you budget; list rates change.
Open weights change the cost shape.
PaddleOCR-VL has open weights. You can run it yourself. Hosted OpenParser is the same model behind an API.
- You own the GPU bill. Self-hosting replaces per-page API charges with compute, storage, and ops staff.
- Ops work stays yours. Queues, retries, autoscaling, model upgrades, and observability do not disappear when you leave a hosted API.
- Model your break-even. GPU mix, utilization, and on-call cost vary by deployment. Use your own capacity plan before comparing self-hosting with a per-page API.
- Hosted path. OpenParser operates the model so you call parse and extract over HTTP. $10 free credits to start, then $1 per 1,000 parsed pages on open-weight hosted.
Where the numbers come from.
Public list prices re-checked 2026-07-24.
$10 free credits; $1 per 1,000 pages on open-weight hosted thereafter. Extraction billed per model tokens.
Detect Document Text $1.50 / 1,000 pages (first 1M). Analyze Document Tables $15 / 1,000 pages (first 1M); Layout included with Tables.
Read $1.50 / 1,000 pages (0–1M). Layout and other prebuilt models $10 / 1,000 pages.
Enterprise Document OCR $1.50 / 1,000 pages (1–5M). Layout Parser $10 / 1,000 pages.
$2 / 1,000 pages synchronous; 50% Batch API discount stated by Mistral.
$4 / 1,000 pages synchronous; 50% Batch API discount stated by Mistral.
$0.015 per credit after the first 15K credits. Standard parse = 1 credit / page → $15 / 1,000 pages. Complex / agentic pages cost more credits.
Pricing questions, short answers.
Cheapest-claim scope, free allotment, lanes, and what this page does not claim.
Is OpenParser the cheapest OCR API?
On the layout-and-tables lane in this public list-price table, OpenParser open-weight hosted OCR is $1 per 1,000 pages after $10 free credits, the lowest rate among the layout rows shown. That is not a claim that OpenParser is cheapest for every OCR job. Plain-text SKUs from AWS, Azure, and Google list at $1.50 and return less structure. Batch discounts, volume tiers, and negotiated contracts can reorder the table.
What does $1 per 1,000 pages cover?
Parsing: turning PDFs and images into normalized typed blocks and markdown. Schema extraction is separate and adds the token cost of the model you pick.
Why split basic-text OCR from layout pricing?
Hyperscalers publish different SKUs for plain text versus tables and layout. Comparing a $1.50 text OCR SKU to a layout parser mixes two jobs. The table keeps those lanes separate so “cheapest” stays scoped to a real workload.
Do you include batch or volume discounts?
No. Figures are synchronous public list rates from the cited vendor pages, dated on this page. Mistral’s published 50% batch discount and hyperscaler volume tiers are noted in sources but not applied in the calculator.
Can I self-host PaddleOCR instead?
Yes. Open weights let you run the model on your GPUs. You then own serving, scaling, queueing, and model upgrades. Self-hosting replaces per-page API charges with compute and ops cost; this page does not invent a break-even month.
Does cheapest list price mean the best stack?
List price alone does not. Output shape still matters: text-and-line-box OCR is a different product from layout, tables, and figures. Use the lanes and caveats on this page before you budget.
Upload a document,
then model the bill.
Upload in Open Studio to see typed blocks and markdown. When you are ready for production volume, create an API key.