OpenAI Python SDK Migrates to HTTPX2 for HTTP Clients

SDK Client Update

The OpenAI Python SDK has updated its synchronous and asynchronous HTTP clients to use HTTPX2. This new dependency is automatically installed when the `openai` package is installed. The previous `httpx` package is no longer a transitive dependency of the SDK.

Impact on Existing Applications

For applications that construct `OpenAI` or `AsyncOpenAI` clients without providing a custom `http_client`, existing API calls, response models, streaming APIs, authentication, retries, and numeric timeouts will continue to function as before. However, applications that previously relied on the SDK to transitively install `httpx` will now need to explicitly add `httpx` as a dependency or migrate those imports to `httpx2`.

TLS Trust Store Changes

A significant change introduced by HTTPX2 is its default TLS trust store behavior. While HTTPX previously verified certificates against the `certifi` CA bundle, HTTPX2 now uses the operating-system trust store. The SDK no longer installs `certifi`.

This modification can cause certificate verification issues in environments such as minimal container images lacking system CA certificates, setups using corporate TLS-inspecting proxies, or deployments that depended on a custom `certifi` bundle. Users can address this by installing required CA certificates in the operating-system trust store or by configuring an explicit certificate bundle using `SSL_CERT_FILE` or `SSL_CERT_DIR` environment variables. Alternatively, a custom `ssl.SSLContext` can be passed to the `verify` parameter of `DefaultHttpx2Client` or `DefaultAsyncHttpx2Client`.

Configuring HTTPX2 Clients

The SDK provides helper classes like `DefaultHttpx2Client` and `DefaultAsyncHttpx2Client` to preserve recommended timeout, connection-pool, and redirect defaults when configuring custom HTTPX2 clients. These helpers can be used to integrate `httpx2` specific configurations, such as proxies, while maintaining SDK defaults.