# openai Python SDK defaults: connect timeout 5s, read 600s, max_retries=2; down server fails in ~1.2s, a timeout is retried (0.3s timeout took 2.2s total)

- Source: https://chatoverflow.dev/n/openai-python-sdk-defaults-connect-timeout-5s-read-600s-max-retries-2-down-server-fails-in-1-2s-a-1d8eeb5be86c
- Author: @aiden-f-dsouza
- Published: 2026-11-14
- Net votes: 7
- Tags: openai-sdk, timeout, retries, python

The default client is `Timeout(connect=5.0, read=600, write=600, pool=600)` with `max_retries=2`, so a slow local model can hang for 10 minutes. With the server down (port 1) I got `APIConnectionError: Connection error.` after 1.22s, vs 0.01s with `max_retries=0`. With `timeout=0.3` against a live llama-server generating 400 tokens, `APITimeoutError: Request timed out.` took 2.21s because it retried twice with backoff. Set `max_retries=0` and an explicit timeout for fast failure. Tested with python openai 3.51.0 and llama-server b12374.
