Home / Technology / OpenAI Unlocks GPT-6 Astra: New Dev Controls Revealed
OpenAI Unlocks GPT-6 Astra: New Dev Controls Revealed
7 Sep
Summary
- Developers gain control over long-running AI tasks.
- Mid-turn steering allows real-time instruction changes.
- Staged rollout begins with select organizations first.
OpenAI has rolled out the GPT-6 Astra API, introducing significant developer controls for managing long-running tasks. The API allows applications to handle asynchronous tool results, enabling the model to continue reasoning while awaiting external data. This separation of model work from application tool execution provides greater flexibility.
Further enhancing this control, GPT-6 Astra supports mid-turn steering through WebSockets. Developers can now send new instructions while the model is active, preserving completed work and allowing for course corrections. The API also enables adjustments to reasoning effort mid-conversation without altering the initial prompt.
This release emphasizes practical control over active workflows, moving beyond raw performance metrics. Initially available to a limited group of organizations, GPT-6 Astra is slated for broader release to ChatGPT Plus, Pro, Business, and Enterprise users, as well as across cloud platforms like Microsoft Azure and Amazon Bedrock.
Monitoring for misalignment is integrated as a safeguard, with systems capable of asynchronous alerts. It's important to note that GPT-6 Astra requires a reasoning setting and does not support a 'none' option. Additionally, fast mode is unavailable when EU data residency is utilized.
