OpenAI Python SDK v3.19.0 Adds GCP Storage Support and References Unreleased 'GPT-Rosalind' Model

What shipped

OpenAI released version 3.19.0 of its official Python SDK on September 22, 2026. The release adds two new features and fixes five bugs, according to the changelog published on GitHub.

Features

The most notable addition is API support for GCP external storage, letting the SDK reference Google Cloud Platform storage locations for API operations (PR #3943). This extends OpenAI's existing external storage integrations, which previously covered other cloud providers, to GCP.

The second feature is a reference to a GPT-Rosalind research model (PR #3940). The changelog entry only adds the model identifier to the client library — it does not include specifications, benchmark scores, pricing, or a context window size. OpenAI has not published a model card, blog post, or announcement for GPT-Rosalind as of this release. The naming pattern is consistent with OpenAI's practice of shipping SDK support for internal or upcoming research models ahead of a public launch, but nothing in this changelog confirms the model is generally available, what it does, or when — or whether — it will be released publicly.

Bug fixes

Five fixes landed in this release:

• Async transform propagation: __api_exclude__ now correctly propagates through _async_transform_recursive, fixing a bug where excluded fields could leak into async request payloads (PR #3324).
• WebSocket omission markers: The client now correctly handles omission markers in queued WebSocket events, addressing a bug that could cause malformed event processing (PR #3944).
• Retry safety: The client now retries only replayable request content, preventing potential double-submission of non-idempotent request bodies on retry (PR #3771).
• aiohttp compatibility: The SDK now tolerates older optional aiohttp installations rather than failing outright (PR #3941).
• Async loop handling: Helper methods now use asyncio.get_running_loop() inside async methods instead of deprecated loop-fetching patterns (PR #3289).

What this means

This is a routine SDK maintenance release, not a model launch. The GCP storage addition is a incremental infrastructure improvement for developers running OpenAI workloads with Google Cloud storage backends. The bug fixes address real reliability issues — particularly the retry-safety fix, which prevents the client from resending non-idempotent requests, and the WebSocket fix, which affects real-time API consumers using the Realtime API.

The GPT-Rosalind reference is the item worth watching. OpenAI has a history of adding model identifiers to its SDK before public announcements, sometimes weeks ahead of a release. Developers should not expect to call this model yet — there's no confirmation it's accessible via the public API, no pricing, and no benchmark data. Treat it as a signal that a research model is in OpenAI's pipeline, not as a product announcement.