Slow import times (due to openai.types)

[jmathes]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• This is an issue with the Python library

Describe the bug

On average, it takes a long time to run import openai on recent hardware and software.

Average of 2 seconds on:

• Windows 11 with
• Intel(R) Core(TM) Ultra 7 165H (3.80 GHz)
• 32.0 GB (31.6 GB usable)

Average of 0.7 seconds on:

• Ubuntu 24.04 HP Z420 workstation
• Intel Xeon E5-1650 v2 x 12
• 64.0 Gib

If that sounds normal, consider the case where a bunch of deps are imported and each one of them takes 2 seconds to be imported. For example, I discovered this issue when trying to use Microsoft's Agentic Framework and Semantic Kernel libraries. Both of those routinely take over 30 seconds to import, spending most of that time importing openai. This renders the libraries useless for CLI or dev utilities and limits them to long-running applications.

This is a near-duplicate of Slow import times #2339, which was fixed, but has reappeared. It seems the new slowdown is caused by expansive internal typing libraries. These can be helpful for developing but are likely irrelevant to the consumer. f

To Reproduce

• (powershell) Measure-Command { python.exe -c "import openai" }
• (*nix) time python.exe -c "import openai"
• python -X importtime -c "import openai"

Code snippets

OS

Windows 11, Ubuntu 24.04

Python version

Python v3.12.8 (on both Windows and Ubuntu)

Library version

openai v2.15.0 (on both Windows and Ubuntu)

[TheCoreMan]

Seeing similar things here. We use pyleak to detect event loop blocking in our app and our CI started failing recently, with the stacktrace pointing towards importlib when trying to import openai.