The ChatGPT GPT Astra API rollout now includes developer controls for long-running work, according to OpenAI‘s current guidance. The documentation says GPT-6 Astra can continue reasoning or call other tools while an application waits for an asynchronous tool result.

That design separates the model’s work from the application’s tool execution. The application still runs the tool and manages the pending result. OpenAI’s documentation says the tool can return its result later using the original call identifier, allowing a developer to keep the workflow connected.

The API also adds mid-turn steering through WebSockets. Developers can send another instruction while GPT-6 Astra is working, and the Responses API can preserve completed work in a continuation. This is aimed at cases where a user changes a requirement or corrects the direction of a task before it finishes.

OpenAI lists a separate control for changing reasoning effort during a conversation. A configuration update can increase or reduce the setting without rewriting the original prompt prefix. The documentation says the updated setting remains in effect until another configuration update replaces it.

The same guidance lists misalignment monitoring as part of the Astra safeguards. OpenAI says its systems can monitor asynchronously and trigger alerts when necessary. The documentation also states that GPT-6 Astra does not support a none reasoning setting and that fast mode is unavailable when EU data residency is used.

These controls make the developer surface different from a single prompt sent to a chatbot. A product team must still execute its own tools, manage pending calls and decide which user instructions can change an active task. The documentation describes the available behavior, but it does not remove the need for application-side permissions and logging.

OpenAI’s product announcement says GPT-6 Astra is rolling out first to a limited group of organizations. It says the model will then reach ChatGPT Plus, Pro, Business and Enterprise users, as well as the OpenAI API, Microsoft Azure and Amazon Bedrock. The announcement also lists the API model identity as gpt-6-astra.

Axios reported before the launch that OpenAI planned to limit Astra’s most advanced cybersecurity features to a small group of testers. That report identified the model’s cyber capability as a reason for the staged approach. The API documentation’s monitoring and steering controls provide a more practical view of how developers will work within that release.

For teams testing the model, the important change is control over an active workflow rather than a new benchmark number. They can wait for tools, steer a live response and adjust reasoning settings through documented interfaces. Availability, limits and safety checks remain part of the rollout and should be confirmed in OpenAI’s current documentation before deployment.