OpenAI quietly showed off its next generation of AI models to stakeholders in Washington, D.C. on August 1. The model family, tentatively called “Astra,” is designed for long-running, multi-agent tasks, a meaningful leap beyond the GPT-5.6 series the company rolled out just weeks earlier. The preview was private, selective, and held in the nation’s capital.
What Astra actually does differently
The GPT-5.6 series, which includes models branded Sol, Terra, and Luna, launched in June and July 2026. Astra builds on that foundation but pivots toward something more ambitious: coordinating multiple AI agents working together on complex, extended tasks. OpenAI hasn’t released benchmarks, performance metrics, or a timeline for public availability. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
