Hi ,
Thanks for reaching out to Microsoft Q&A.
This might be due to...
- High Resource Demand:
- The High Resolution and Style flags increase the computational load significantly.
- When used together on large documents, they can exceed internal processing limits, especially on memory or time constraints.
- Markdown Output Serialization:
- The Markdown output format requires more post-processing compared to raw JSON or plain text.
- This gets compounded when style and layout metadata are involved, which increases the size and complexity of the response.
- Backend Timeout or Resource Limit:
- The model likely hits an internal timeout or quota (ex: memory, inference time), triggering a 500 error rather than a graceful fallback.
Workarounds:
- Use only one flag at a time (as you already noted): Either use High Resolution or Style, but not both together.
- Switch back to plain text or JSON output for large documents.
- Split large PDFs into smaller chunks (10 to 15 pages per request).
- If Markdown is essential, consider post-processing the output yourself: Get raw JSON > parse it locally > convert to Markdown with your own logic.
Currently, this limitation is not well-documented. However, it has been observed by multiple users in real-world usage and GitHub issues. Internally, it likely relates to system-level resource caps.
Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the reply was helpful. This will be benefitting other community members who face the same issue.