OpenAI builds Astra, an AI model that can work long hours or days

OpenAI is working on a new family of AI models, called Astra. It’s an AI that can work on a single task for hours or even days. This model can manage multiple agents at once, according to three people familiar with the matter (via The Information). These agents are like employees of the AI, and they’re intended for complex tasks such as mathematics, science, and long-term research projects.

Astra has been shown to policymakers in Washington this week by Sam Altman. It’s the first OpenAI model to go through a new security check initiated by the US government. The final name and launch date are still unknown. It might be released as GPT-6 or GPT 5.7.

OpenAI has announced that it wants to create an autonomous AI researcher by March 2028 and a research intern by September 2026. Astra is another step towards these goals. Instead of a single superchatbot, we’ll have a team of specialized agents working together.

This is a paradigm shift. We’re going from an informational AI to a taskmaster AI. With Astra, there won’t be any difference between AI assistants and AI employees. The tech industry will be the first to notice.