OpenAI added computer use to the Agents API on September 29. An agent can now work inside an OpenAI-hosted browser to navigate sites and interact with web interfaces instead of relying only on APIs. OpenAI's computer-use guide
The application still owns important permissions. It has to handle access requests for each new website origin and manage sign-in when a task needs an account. OpenAI also warns that approving a site does not create confirmation before every individual browser action, so consequential workflows still need their own controls.
The same day's API changelog added Multi-agent support in beta for GPT-6.1 Sol. In a Responses API request, the model can delegate parts of a task to subagents rather than carrying the whole job in one thread.
The multi-agent beta is tied to GPT-6.1 Sol; the announcement does not extend it to every OpenAI model.