OpenAI Python SDK v3.22.0 Adds Computer Use to Beta Agents API

What shipped

OpenAI released version 3.22.0 of its official openai-python client library on GitHub. The single feature listed in the release notes: computer use added to beta agents (PR #3989, commit f5dddb5).

The release tag lists a date of 2026-09-29, which appears in the changelog metadata generated by OpenAI's SDK release tooling. No further details, migration notes, or breaking changes are documented in the release body beyond the single feature entry.

What this actually is

This is a client-library update, not a new model or a new API capability announced independently — it exposes computer use functionality inside the SDK's beta agents module. Computer use is OpenAI's existing capability that allows a model to control a computer interface (mouse clicks, keyboard input, screenshots) to complete tasks, previously accessible via the Responses API and Operator-style tooling. This release wires that capability into the Python SDK's beta agents abstraction, making it easier for developers building agentic workflows with the openai-python library to invoke computer use without manually constructing raw API calls.

No new model weights, benchmark scores, context window changes, or pricing updates accompany this release. It is purely a developer-facing SDK change.

Why it matters for developers

Developers who use OpenAI's beta agents framework in Python can now integrate computer-use actions directly through the SDK's agent objects rather than hand-rolling requests against the underlying API. This lowers the integration burden for anyone building autonomous or semi-autonomous agents that need to interact with GUIs, browsers, or desktop applications.

As with all releases tagged "beta," the agents API and its computer-use integration remain subject to change. OpenAI has not published stability guarantees or a timeline for graduating this out of beta.

What this means

This is routine SDK maintenance, not a model release. It signals that OpenAI continues to invest engineering effort in making its agent-building tools more capable in code, following the broader industry push toward agentic AI products. Teams already building on the beta agents API should update to v3.22.0 to access computer use natively; teams not using agents or computer use can treat this as a no-op update. No pricing, benchmark, or capability claims require independent verification here — the change is confined to the SDK layer, with computer use itself an already-existing OpenAI capability now surfaced through a new interface.