OpenAI Astra model revealed, focuses on multi-agent long-horizon collaboration
The Information reports OpenAI is testing a new model codenamed Astra. Altman demonstrated it to regulators in Washington. Astra may become one of the first frontier systems reviewed under the new US government framework.
On August 1, The Information reported that OpenAI is testing a new model series codenamed Astra. OpenAI CEO Sam Altman this week demonstrated the system in closed-door sessions to several US senators and regulators in Washington.
Astra's core capability is supporting multiple AI agents working in parallel within a unified task framework, sustaining complex long-horizon tasks. The model assisted in disproving the 80-year-old Erdős unit distance conjecture during internal testing, which the industry views as a key OpenAI move in multi-agent collaboration.
Regarding naming, Astra is currently an internal codename. OpenAI has not decided whether to release it as standalone GPT-6 or fold it into the GPT-5 series as GPT-5.7. Astra will sit alongside the Sol, Terra, and Luna model tiers.
The report noted that Astra is expected to be one of the first frontier models reviewed by federal agencies under the Trump administration's new framework before public release. Executive Order 14409 issued on June 2 has established the framework, allowing eligible models to be opened to federal evaluation up to 30 days before release to other trusted partners.