OpenAI’s Next Frontier: The ‘Astra’ Model Series for Sustained AI Teamwork
OpenAI is preparing to unveil a new family of AI models, internally codenamed “Astra,” with a focus on enabling multiple AI agents to collaborate over extended periods, according to a report from The Information citing people familiar with the matter.
Beyond Single Tasks: Coordinating Agents for Complex Challenges
The key advancement of Astra lies in its designed capability to manage long-horizon tasks. Instead of performing isolated actions, multiple AI agents powered by Astra could work together coherently for weeks or months, akin to a project team.
This opens up applications in areas like managing large-scale software development cycles or, more ambitiously, tackling advanced, unsolved mathematical problems. OpenAI has positioned Astra as a tool for solving particularly difficult, multi-step challenges.
Demonstrations and Regulatory Alignment
OpenAI founder Sam Altman reportedly demonstrated Astra’s capabilities to policymakers and regulators in Washington, D.C. this week, signaling an effort to engage with authorities ahead of a potential launch.
The release may coincide with new U.S. government frameworks for advanced AI. Sources suggest Astra could be among the first models submitted under upcoming rules requiring federal review before public deployment.
Branding and Capabilities Showcase
It remains undecided whether Astra will be branded as GPT-6 or as an extension of the GPT-5 series (e.g., GPT-5.7). It would represent a new class of models following OpenAI’s Sol, Terra, and Luna.
In the near term, OpenAI is said to be preparing a research paper detailing how its technology solved ten previously open mathematical problems, aiming to showcase its state-of-the-art reasoning abilities.
No official launch date for Astra has been announced, but its development highlights OpenAI’s push toward more persistent, collaborative AI systems.