I cannot download large batch output files (>100MB) when trying to retrieve batch processing results via the Files API. Downloads consistently fail mid-transfer with connection errors.
Timeline: This worked fine a few weeks ago, but has been failing consistently for several days now.
Details:
-
File size: 242 MB
-
Error: Connection drops after 10-30 MB downloaded
-
Small files (<10MB): Download successfully
-
Large files (>100MB): Consistently fail
Errors encountered:
-
APIConnectionError: Connection error -
504 Gateway Time-out -
HTTP/2 stream error: INTERNAL_ERROR
Attempted workarounds:
-
OpenAI Python SDK with various settings
-
Chunked downloads with HTTP Range headers → results in corrupted data with duplicate truncated lines
-
curl with retry logic and HTTP/1.1
-
Different networks/VPN
Initial connection succeeds (HTTP 200 OK) but drops partway through transfer, suggesting a timeout or streaming issue on OpenAI’s infrastructure.
Has anyone else experienced this? Is there a known workaround, or is this an OpenAI infrastructure issue that needs to be addressed?