Sol, Terra, and Luna are OpenAI's next generation frontier models. A small group of partners can see them first, the public cannot, and the company is publicly resisting the arrangement becoming standard.
OpenAI has unveiled three new flagship models, Sol, Terra, and Luna, as part of what the company is calling the GPT-5.6 family, and is keeping them out of public hands at the US government's request. The models are real, named, and documented on OpenAI's own help site. They are also, for now, invisible to anyone outside a small group of "trusted partners" the company has selected, plus the federal reviewers sitting in front of them.
The arrangement is the story, not the model specs. OpenAI is the first major US AI lab to publicly submit its newest frontier systems for a pre-release safety review by Washington, and it is doing so while arguing, on the record, that this should not be how frontier AI gets released from now on.
According to OpenAI's announcement page and reporting summarized by LatestLY, Sol is priced similarly to GPT-5.5 while offering improved performance and a new reasoning mode. Terra and Luna round out the trio, and OpenAI has published a dedicated preview article for all three on its help center. These are not rumored models. They are shipping internally, and the public knows about them.
What the public does not get is access. OpenAI says it is honoring a request from US authorities for an early safety review before general release, a framing corroborated by The Verge, TechCrunch, and CNBC. The company has not named the requesting agency, has not disclosed what the review is actually testing for, and has not said who its "trusted partners" are. One industry roundup, MLQ.ai, calls this the "first-ever US government-gated AI rollout," but that phrasing is editorial emphasis, not OpenAI's own language.
The tension is in what OpenAI is saying next. The company has told reporters, including TechCrunch, that the restricted rollout "shouldn't be the norm," arguing that routine pre-release government review could slow innovation and reduce access for developers, businesses, and security researchers who depend on cutting-edge systems. OpenAI is, in other words, complying with a request it is simultaneously warning against accepting as a template.
That contradiction is what makes this more than a product launch. Wire coverage treats the story as a delayed release with a federal footnote. The substance is that a frontier lab has agreed to a quiet, pre-publication review of its most capable systems, one that no other US AI lab has publicly accepted, and is now using its platform to argue against the precedent that creates. Whether the arrangement sticks depends on questions OpenAI has not answered: which agency is reviewing, what the review actually evaluates, who the partners are, and whether other labs will be asked to do the same thing next time a generation ships.
For now, Sol, Terra, and Luna are visible on OpenAI's own pages and invisible to almost everyone else. The government's window is open. The public's is not.