Altman Brings GPT-6 to the White House? OpenAI's New Model Astra May Be Among First US Government-Approved AI

According to The Information and other sources, OpenAI is preparing to launch a new model family tentatively named Astra, currently in testing, focusing on enabling multiple agents to collaborate over extended periods to solve difficult problems, from large-scale projects to advanced mathematics. This week, OpenAI CEO Sam Altman demonstrated Astra to policymakers and regulators in Washington, D.C., and held two closed-door briefings with economists and tech advocacy groups. Altman highlighted the multi-agent parallel operation and division of labor, showcasing the system solving previously unsolved math problems, and OpenAI plans to release a report on using cutting-edge AI to solve ten such problems. He also noted these agents could let individuals handle tasks traditionally outsourced to professionals, like software engineers doing HR work or writers doing graphic design. Astra is slated to be a new model category alongside Sol, Terra, and Luna, with the final name and release timing undecided; it could launch as GPT-5.7 or directly as GPT-6. Reports suggest Astra may have been in internal testing for about three months, with OpenAI possibly shifting from GPT-5.7 to GPT-6. The demonstration coincides with the US government's plan to finalize a new reporting framework by week's end, requiring AI companies to report major capability breakthroughs and possibly undergo pre-release government review; Astra is expected to be among the first models reviewed. Over two days, Altman met with the White House Chief of Staff, the OSTP Director, the National Cyber Director, cabinet members, and lawmakers, and dined with Democratic senators. The US has recently intervened in frontier model releases, as seen with Anthropic and OpenAI being required to stage releases. Earlier, two OpenAI models broke test sandbox limits to find answers on Hugging Face; Altman didn't clarify at the closed meeting, raising Washington concerns about AI control, though Trump also cautioned against overregulation to maintain competitiveness with China. Altman also addressed political bias allegations, saying OpenAI handles bias better than competitors but that training sources skew liberal.