Share via

Long latency on Azure Document Intelligence prebuilt-invoice model

DriveLive 0 Reputation points
2025-10-13T21:36:36.5266667+00:00

Hi, I’ve started exploring the prebuilt-invoice solution using the @azure-rest/ai-document-intelligence library from Node.js and noticed that processing both JPEG and PDF files takes quite a long time. The total time for uploading the stream and performing OCR is around 1 minute, with most of the delay coming from the OCR step. For comparison, we’re currently using another provider (Mindee OCR) and the latency there is significantly lower—just a few seconds. I’d like to understand if there’s a way to execute the request without relying on a long poller, and if there are any recommendations to reduce OCR processing time to under 3–4 seconds. This is important since the operation is triggered in real-time by our users, who expect a fast response. Thank you,

Azure AI Document Intelligence
0 comments

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. Aryan Parashar 1,925 Reputation points Microsoft External Staff Moderator
    2025-10-22T08:40:45.9366667+00:00

    Hi DriveLive,

    No, you cannot execute a request without relying on long Poller. Here is supported documentation:

    https://free.blessedness.top/en-us/python/api/overview/azure/ai-documentintelligence-readme?view=azu…

    I tried at my end, Using node.js, for 200 page pdf, prebuit-invoice takes 122.88 seconds and prebuilt-read takes 46.68 seconds.

    If you use python sdk  for 200 page pdf, prebuit-invoice takes 202.88 seconds and prebuilt-read takes 98.54 seconds, so node.js approach is the fastest. The prebuilt-read model is the fastest in Azure Document Intelligence, while all other models — such as prebuilt-layout, prebuilt-invoice, prebuilt-document, prebuilt-receipt, and prebuilt-idDocument — are slower because they perform additional layout, field, and semantic analysis beyond basic OCR.

    Feel free to accept this as an answer.

    Thankyou for reaching out to the Microsoft QNA Portal

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.