OpenAI has been cleared to release its most advanced model widely, after the US government signed off on a broader rollout of GPT-5.6 that had been held back for weeks under Washington’s new oversight regime for frontier AI.
Until now the model had been available only through a restricted preview to about 20 partners whose names were individually approved by the US government. That arrangement, the first of its kind for an American frontier model, is what the wider release now supersedes.
The sign-off followed additional testing by the Commerce Department’s Center for AI Standards and Innovation, the body set up to vet advanced systems. OpenAI sent technical experts to Washington to answer the agency’s questions, according to Axios.
GPT-5.6 is a three-tier family rather than a single model. Sol is the flagship, Terra a lower-cost mid-tier option, and Luna the fastest and cheapest of the three.
OpenAI has described Sol as strong at coding, biology and cybersecurity, and paired it with a** “max reasoning effort”** mode that gives the model more time to work through hard problems. Those same capabilities, particularly in biology and cyber, are part of why the government wanted a closer look before a wide release.
The tiering is a commercial choice as much as a technical one. Terra is aimed at everyday enterprise workloads where cost matters more than raw capability, while Luna is built for high-volume tasks that need speed above all, a split that lets OpenAI charge very different prices across the same family.
The preview it now supersedes was unusually tight. For weeks GPT-5.6 was available only to a short list of organisations whose identities OpenAI had shared with the government, the first time an American lab had gated a frontier model behind a state-approved roster.
The review sits inside a framework the Trump administration established on 2 June, which introduced a voluntary pre-release check for the most capable models. The GPT-5.6 case went further than that, moving from voluntary review to a government-managed access list, a step OpenAI had agreed to only after being asked to slow the launch.
OpenAI has made clear it is uneasy with the precedent. The company said it does not believe this kind of government access process should become the long-term default, while agreeing to take part this time.
The discomfort is not hard to understand. A government that can gate a launch can also stop one, a power the administration has already used elsewhere in the sector by ordering Anthropic to shut down two models.
For OpenAI, the commercial stakes of the delay were real. Every week that GPT-5.6 stayed inside a 20-partner preview was a week rivals could court the enterprise customers it wanted to reach with the new tiers. The company now expects to widen access to GPT-5.6 within days, building on the base it laid with GPT-5.5 earlier in the year. It has said all three tiers will become generally available in the coming weeks, though it has not fixed a public date.
OpenAI is not the only lab inside the new regime. The same framework covers its rivals, which means the way this rollout has gone is likely to shape how the next frontier model from any US company reaches the public.
What the episode establishes is less a product timeline than a template. For the first time, a leading US lab has released a frontier model on the government’s schedule rather than its own, and both sides now have to decide whether that was a one-off or the shape of things to come.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.