- Lead. OpenAI launched a restricted preview of GPT-5.6—a three-tier model family named Sol, Terra, and Luna—on June 26, limiting access to a small group of government-approved partners at the explicit request of the Trump administration.
- Fact. OpenAI confirmed the restriction publicly while making clear it viewed it as temporary: “We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises.” Broader access is expected within weeks.
- Stake. This is the first time a US presidential administration has asked an American AI lab to pre-screen domestic access to a new frontier model before public release—establishing a precedent for executive-branch involvement in how AI capabilities reach the market.
Sol, Terra, Luna: the GPT-5.6 family
GPT-5.6 is structured as three models differentiated by price and capability. Sol, the flagship tier, is priced at $5 per million input tokens and $30 per million output tokens via the API. Terra, the balanced tier, is priced at $2.50 and $15. Luna, the fast low-cost option, is priced at $1 and $6. The pricing places the family at the high end of frontier-model API rates—above the current GPT-4o pricing but below OpenAI’s most expensive reasoning models—while offering customers a range of options based on their latency and cost tolerance.
Access to all three models at launch was restricted to a “small group of trusted partners” whose participation was shared with the government. OpenAI characterised the limited preview as a “short-term step,” with a broader rollout planned for coming weeks.
The administration’s request
The Trump administration made the access request on approximately June 25, the day before launch. OpenAI complied. The company indicated it was simultaneously working with the administration on a new cybersecurity executive order framework and seeking to develop a “repeatable process for future model releases”—suggesting the June 26 arrangement was conceived as a template rather than a one-off accommodation.
Dean Ball, a former White House AI adviser who is joining OpenAI, was named in coverage of the episode. His transition from the executive branch to the lab reflects the increasingly direct relationship between the administration and frontier AI developers—a relationship that has also generated friction, including the subpoenas issued by 42 state attorneys general against OpenAI in the weeks prior.
Precedent and the undisclosed list
The US government has previously exercised control over AI capabilities through export controls, restricting which chips and models can leave the country. It has not previously asked a US lab to restrict access to a domestic commercial model release within the United States before a public launch. The June 26 arrangement moves that authority inward: from border control to launch control.
OpenAI’s public statement was notable for the tension it held simultaneously—accepting the restriction as appropriate now while asserting it “shouldn’t be the norm.” TechCrunch, which first reported the details, noted that the roughly 20 companies approved for the preview hold a structural first-mover advantage in a market where model capability is a direct competitive input. The identities of those companies—and the criteria by which the administration approved them—have not been disclosed by OpenAI or the White House.