Overview
- OpenAI started a restricted deployment of its GPT-5.6 family to a small group of trusted, U.S.-based partners on June 26, with the company saying it will add international partners in the coming week.
- CEO Sam Altman told staff that the U.S. government will approve access client by client during this limited launch, giving federal authorities direct oversight of who may use the new models.
- GPT-5.6 includes three models—Sol (flagship with advances in software engineering, autonomous computer use, scientific research and cybersecurity), Terra (balanced) and Luna (fast, low-cost)—and OpenAI says Terra will cost about half as much as GPT-5.5.
- Anthropic was ordered on June 12 to cut access to its Mythos and Fable models for foreign nationals and remained offline for weeks before the government narrowly reauthorized Mythos for a restricted set of U.S. partners, leaving broader foreign access still limited.
- The U.S. controls and OpenAI’s policy allowing partner employees abroad to use GPT-5.6 in most non-sanctioned countries heighten digital-sovereignty concerns, increase incentives for open-source or non-U.S. alternatives, and could reshape who builds and deploys frontier AI.