OpenAI is reportedly developing the Astra model family for long-running agentic tasks
OpenAI is reportedly developing a new model family, Astra, for tasks lasting hours to days, coordinating multiple agents. Sam Altman demonstrated it to politicians in Washington, and the model is being tested as the first case under the new US regulatory framework.
According to The Information, which cites three people familiar with the plans of the company, OpenAI is developing a new model family with the working name Astra. The goal is a system significantly more capable at long-running tasks than anything the company has released so far – it is intended to coordinate multiple agents simultaneously to solve particularly challenging problems, such as complex projects or advanced mathematics. CEO Sam Altman demonstrated the Astra model to politicians and regulators in Washington, D.C. this week.
Astra is expected to form a new model class at OpenAI alongside the existing Sol, Terra and Luna families. Whether it will be released as GPT-6 or as a variant within the GPT-5 series (for example, GPT 5.7) has not yet been decided, nor has the release date. The model is already being tested and, according to the source, is expected to be the first to go through the planned new regulatory framework of the Trump administration – this would require AI models to be submitted to the federal government before public release. According to the article, the administration is expected to finalize this framework by the end of this week.
The source highlights the open question of whether Astra can prevent errors from accumulating during long-running tasks and correct itself when the process goes off course as the context grows – according to the article, this remains the main weakness of current agentic systems. Systems with multiple agents may also perform worse on tightly interdependent tasks, such as planning, because coordination overhead and accumulating errors may outweigh the benefits.
The plan follows earlier comments by the chief scientist at OpenAI, Jakub Pachocki, who said on the company podcast last summer that the goal is to create systems capable of working on a problem for hours to days. According to the article, the company plans to have a fully autonomous AI researcher capable of independently leading research projects by March 2028, and is expected to have a system at the level of a research intern as early as September this year. According to the article, Astra could become that system.
Why it matters
If Astra is confirmed and receives regulatory approval, it will be the first test of the new US framework requiring advanced AI models to be submitted to the government before public release – this may affect the pace at which companies such as OpenAI deploy their most powerful models. For developers and companies working with agentic systems, the reported risk of accumulating errors in long-running tasks and the possible drop in performance on tightly interdependent tasks when coordinating multiple agents are particularly relevant.
Two audiences, two different impacts
What this means
For individuals
People working with AI agents should bear in mind that, according to the source, models for long-running tasks still struggle with accumulating errors and coordination overhead, which limits their reliability on complex, interdependent tasks.
More practical updates →For a business
Companies planning to deploy advanced agentic models from OpenAI should monitor the emerging US regulatory framework requiring models to be submitted to the government before public release, as it may affect the availability and timing of future versions.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.