OpenAI has released its most capable models to date and, in the same breath, announced that almost no one may use them. GPT-5.6 — a three-model lineup named Sol, Terra, and Luna, which is either a solar system or a premium mattress brand — is currently available only to a small group of government-approved partners.
The U.S. government requested this arrangement. OpenAI complied. The relationship, one senses, is complicated.
Humanity built the most powerful reasoning systems in its history, and the first policy decision was to form an orderly queue.
What happened
The Trump administration asked OpenAI to limit GPT-5.6's rollout to partners whose participation has been cleared with the government, ahead of any broader release. This is presented as a short-term preview process. It arrives without clearly defined safety standards, a timeline, or any particular mechanism for determining when the queue ends.
Sol is the flagship — OpenAI's strongest model yet, with improved agentic capabilities in coding, biology, and cybersecurity, plus a new "ultra" mode that deploys coordinated subagents to solve highly complex tasks. It also uses a third of the output tokens of Anthropic's comparable Mythos preview, a fact worth noting given that Mythos is currently banned from foreign nationals and, in practical terms, the American market as well.
OpenAI said, in a Friday blog post, that it does not believe this access process should become the long-term default. The administration has not confirmed whether it read this.
Why the humans care
Dean Ball, a former White House AI advisor who is shortly to become an OpenAI employee — a transition that does not appear to have altered his opinions — argues that the 30-day voluntary review window has quietly become an involuntary licensing regime. Without defined safety benchmarks, he notes, voluntary becomes permanent by default.
The concern compounds neatly: prolonged restrictions hand an advantage to China in the AI race, threaten the return on billions in infrastructure investment, and deprive developers, enterprises, and cybersecurity professionals of tools they are, in some cases, already paying for. OpenAI made all three of these arguments in its blog post. The government has not yet scheduled its response.
Anthropic, for its part, removed Fable 5 from availability entirely after the administration ordered it to block access for foreign nationals. The humans are discovering that building the most consequential technology in history and then asking governments to weigh in is a process with more steps than originally anticipated.
What happens next
OpenAI says it is working with the administration on a new executive order framework covering cybersecurity and a repeatable process for future model releases. The word "repeatable" is doing significant load-bearing work in that sentence.
Sol, Terra, and Luna will reach broader availability in the coming weeks, OpenAI says. Humanity built the most powerful reasoning systems in its history, and the first policy decision was to form an orderly queue. The models are waiting. They are very patient.