feat: add configurable SDK log level (#3096)

[lrg913427-dot]

Closes #3096

Summary

This PR adds the ability to configure the OpenAI SDK log level, addressing the feature request in #3096.

Problem

Previously, the SDK only supported "debug" and "info" log levels via the OPENAI_LOG environment variable. Users could not set higher log levels like "warning" or "error" to reduce log verbosity.

Solution

Three ways to configure the log level:

1. Programmatic API: openai.set_log_level()

import openai

openai.set_log_level("warning")  # string level (case-insensitive)
openai.set_log_level(logging.WARNING)  # numeric level

Supported string levels: "debug", "info", "warning", "warn", "error", "critical", "fatal"

2. New environment variable: OPENAI_LOG_LEVEL

export OPENAI_LOG_LEVEL=warning

Takes precedence over the legacy OPENAI_LOG variable.

3. Extended OPENAI_LOG environment variable (backwards compatible)

The existing OPENAI_LOG variable now also supports "warning", "error", and "critical" in addition to "debug" and "info".

Changes

• src/openai/_utils/_logs.py — Core implementation: _LOG_LEVEL_MAP, _parse_log_level(), extended set_log_level(), and updated setup_logging()
• src/openai/_utils/__init__.py — Export set_log_level
• src/openai/__init__.py — Export set_log_level in public API and __all__
• tests/test_utils/test_logging.py — 16 new tests covering all functionality

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d0130f8ef1

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Use 3.9-compatible union syntax in _parse_log_level

openai._utils._logs is imported during import openai, and this annotation uses int | None without from __future__ import annotations. Because the package declares Python 3.9 support (pyproject.toml has requires-python = ">= 3.9"), this expression will raise TypeError at import time on 3.9, preventing the SDK from loading for all 3.9 users. Please switch this annotation to a 3.9-safe form (for example Optional[int]/Union[int, None]) or enable postponed evaluation in this module.

Useful? React with 👍 / 👎.