Create Request

Submit an inference request for a supported document with up to 2,000 pages. Upon receiving the request, the API immediately returns a Request ID. The input file is divided into batches of 10 pages each, and inference is performed on each batch.

---

Retrieve Results

Retrieve inference results by request ID and check the status of each batch in real-time.

How to get your parsed results:

1. Poll this endpoint until the top-level status is completed (or track each batches[].status individually).
2. For each completed batch, download its result from batches[].download_url.
3. Each download is a JSON object in the same format as the synchronous Document Parse response — elements, content (html/markdown/text), coordinates, usage, and so on — covering the pages from that batch's start_page to end_page.
4. Concatenate the batches in id order to reconstruct the full document.

Each download_url is a pre-signed link valid for 15 minutes; re-fetch this endpoint to obtain fresh URLs. Results remain available for 30 days from the time of the request.

---

List Requests

Retrieve a list of all your document parsing async requests with their status and timestamps.

---

Delete Inference Results

Permanently delete a request's inference results. Any download_url issued earlier stops working.

Only finished requests can be deleted. A request that is still processing is rejected with request_in_progress — retry once it completes.

Deleting an already deleted request succeeds again, so the call is safe to retry.

Delete a single request

Delete multiple requests

Send up to 100 request IDs in one call. Duplicate IDs are collapsed, so the results array can be shorter than the list you sent.

---

Handling Errors

In Document Parse Async APIs, errors can occur in three different scenarios: