Niches · Document parsing & extraction
Process a large batch at a known price and time
Two or three companies' docs show a full fit.
What counts as done
An async job or batch endpoint that takes the whole set, reports per-document status and the billed tier and pages, ends every job in success or a terminal error, and posts a webhook when done.
Measured, for our own tests: pages per minute and cost per 1,000 pages at that volume. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: process these 5,000 PDFs
, run OCR on the whole archive overnight
, how much will it cost to parse this corpus
Does the job 2
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Extend API + MCP B CrowdView Inc. dba Extend | Batch endpoints take 1 to 1,000 files, move PENDING to PROCESSED, FAILED or CANCELLED, notify by webhook, and each run reports metrics.pageCount; credit rates per mode are published. source |
| Reducto API + MCP B Reducto, Inc. | Async jobs with Svix or direct webhooks, a discounted batch queue, and usage.num_pages and credits on every response. source |
Does the job, with a limit 14
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Adobe PDF Services / PDF Extract API C Adobe Inc. | Each document is an async job with a CALLBACK notifier and done or failed states, but there is no batch endpoint, the free tier allows 25 requests a minute, and paid per-page price goes through sales. source |
| Amazon Textract BB Amazon Web Services, Inc. | Start operations run async on S3 files up to 3,000 pages and 500 MB, notify an SNS topic with SUCCEEDED or FAILED and take a ClientRequestToken, but each file is its own job, quotas default to low rates, and responses report no billed pages. source |
| Azure Document Intelligence B Microsoft Corporation | analyzeBatch takes up to 10,000 blob files in one request and keeps status 24 hours, but no webhook is documented, and acc-44 saw a job stay running forever with no terminal error. source |
| Datalab B Endless Labs, Inc. | Async jobs with a check URL and webhooks, 7,000 pages and 200 MB per request, 15,000 pages in flight and published per-page prices, but a batch is parallel single requests with no batch endpoint, and billed pages per response were not confirmed. source |
| Docling B Docling Project (LF AI & Data Foundation) | docling-serve takes several sources per async request, reports task_status pending, started, success or failure, and posts callbacks; throughput and cost depend on the user's hardware, and nothing reports per-page billing because none applies. source |
| Google Cloud Document AI BB Google LLC | batchProcess takes Cloud Storage folders and reports state and individualProcessStatuses per document, with prices per 1,000 pages published, but there is no webhook, only operation polling, and billed pages are not in the metadata. source |
| Invofox D Invofox, Inc. (Delaware), with a registered branch in Spain | ZIP and multi-file imports with webhooks and file states including rejected with an error code, but page limits per file and rate limits are not published, and no per-page price is public. source |
| LandingAI Agentic Document Extraction BB LandingAI, Inc. | Parse and Extract Jobs run on a half-rate standard tier with a 6-hour timeout and signed webhooks, and metadata reports billing, but each document is its own job with no batch endpoint, and page count is capped by the plan's per-minute limit. source |
| LlamaParse API + MCP B LlamaIndex, Inc. | Async jobs with webhook_configurations and per-tier credit prices, but acc-32 found jobs billed at a higher tier than requested and acc-33 a job running over ten minutes for two pages. source |
| Mindee API C Mindee SAS (Mindee Inc. outside the Eurozone) | Async enqueue with webhooks and documented limits (100 MB, no page limit on paid plans), but there is no batch endpoint, and credits per page are not reported per response in what was read. source |
| Mistral OCR API B Mistral AI (RCS Paris 952 418 325) | OCR 4.1 supports the Batch API at a 50% discount, with succeeded and failed request counts per job, but no webhook is documented, and billed pages per document were not seen. source |
| Nanonets API + MCP E Nano Net Technologies Inc. | Async parse and extract take up to 50 inputs per call, a batch endpoint and a credit-usage endpoint exist, but no webhook is in the reference and no rate limits are published. source |
| OpenDocRouter B LlamaIndex, Inc. | Async mode takes up to 500 pages with completed, partial or failed status and a charge_usd per page, but each request is one document, there is no webhook, and only 5 async requests run at once. source |
| Unstructured API + MCP E Unstructured Technologies, Inc. | Long jobs return 202 and can be listed, polled, streamed or cancelled to a terminal state, but there is no batch endpoint or webhook in the docs index. source |
Couldn't check 1
The docs were closed to us, or silent. These carry no mark either way.
| Listing | Why |
|---|---|
| Veryfi API + MCP C Veryfi, Inc. | No batch or webhook detail in the SDK README; the API docs were not read. Pricing is per document of up to 15 pages, per the listing. |
Doesn't do this job 3
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| Clairic BB Cloudflare, Inc. | The model page documents only a synchronous run call, with no batch endpoint, job status or webhook. source |
| Nemotron 3.5 Parse B NVIDIA Corporation | Weights served by vLLM one page per request; no job, batch or webhook layer. NVIDIA reports 14.6 pages a second in double pass on one H200. source |
| Xiaomi-OCR-0 C Xiaomi Corporation | Open weights served by vLLM or SGLang, one image per request; no job, batch or webhook layer. source |
6 listings in the category were added after this reading and aren't marked yet.
What stops agents finishing
- job stays running forever with no terminal error
- billed tier differs from the tier requested
- quota error on an unused key
Around it
Before: file-storage.backup-archive. After: vector-search.ingest-sync, data.local-files.
Where this comes from
Weight 13 of 100 in Document parsing & extraction: 8 of 46 accounts; billing at the wrong tier, jobs that never finish, slow parses and price-per-1,000-page comparisons
The same as data: /api/v1/use-cases/document-extraction.bulk-batch.json, or the MCP tool get_use_case with id document-extraction.bulk-batch. Other jobs in this category: /api/v1/niches?category=document-extraction.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.