Page Extract API
Turn PDFs into JSON that fits your schema.
Send a document and a JSON Schema; get back the data, with the page each row is on. Long filings, scans and tables of thousands of rows included.
renaissance_13f.pdf
{
"holdings": [
{
"issuer": "10X GENOMICS INC",
"value": 21415529
},
{
"issuer": "111 INC",
"value": 84851
},
… 3,061 more
]
}
The most accurate on ExtractBench, for less
Value F1 over ExtractBench's 370 documents, from one page to 192, and what each method costs a page.
| Method | Overall | Short | Medium | Long | A page |
|---|---|---|---|---|---|
| Page Extract pro | 96.80 | 97.46 | 95.48 | 94.86 | 5¢ |
| Page Extract standard | 96.37 | 96.98 | 94.99 | 95.46 | 3¢ |
| LlamaExtract Agentic Plus | 96.17 | 97.02 | 94.30 | 94.65 | 7.5¢ |
| Reducto Deep Extract | 96.08 | 96.84 | 94.80 | 92.85 | 6.3¢ |
| Pulse (Effort) | 96.07 | 96.83 | 94.70 | 93.24 | 10.5¢ |
| LlamaExtract Agentic | 95.65 | 96.06 | 94.79 | 94.73 | 3.1¢ |
| Codex (GPT-6.1 Sol) | 95.31 | 97.01 | 92.96 | 85.46 | 7.1¢ |
Page Extract's rows: our runs of November 8, 2026, with prompts tuned on the benchmark's documents, priced at what we charge. The other rows are from ExtractBench's leaderboard.
How it works
- Upload once. Your PDF goes to Page Search, which reads every page, scans and charts included. The same document serves searches and extractions.
- Describe what you want. A JSON Schema with a description for each field. Lists of any length, nested objects, and values that may be missing.
- An agent reads the document. It looks at every page, writes code to collect the rows, and checks them against the pages before it answers.
curl https://api.extract.pageindex.ai/v1/extract \
-H "Authorization: Bearer $PAGEINDEX_SEARCH_API_KEY" \
-H "Content-Type: application/json" \
-d '{"document": "doc_60a07bcef3094b68806070f2",
"configuration": {
"data_schema": {"type": "object", "properties": {
"holdings": {"type": "array", "items": {"type": "object", "properties": {
"issuer": {"type": "string"}, "shares": {"type": "number"}, "value_usd": {"type": "number"}}}}}},
"tier": "standard",
"cite_sources": true}}'{
"id": "ext_…",
"status": "COMPLETED",
"extract_result": {"holdings": [{"issuer": "APPLE INC", "shares": 1250000, "value_usd": 283912500}, …]},
"extract_metadata": {"field_metadata": {"/holdings/0": {"page": 4}, …}},
"usage": {"pages": 81, "credits": 2.43}
}Pay for the pages you extract
Paid when a job completes; a cancelled or failed job is free. The same credit as Page Search, and $5 of it free to start.
Standard
$0.03 / page
For most documents.
Pro
$0.05 / page
A stronger model, for the hardest layouts.