Axios reported on Tuesday that the Commerce Department’s Center for AI Standards and Innovation, the office that has spent the last five weeks poking at GPT-5.6 behind a security clearance, has finished its review and cleared OpenAI to widen access beyond the 20-partner preview list it has been restricted to since early June. That preview list, which we covered on July 4, was the first time a US frontier model had been gated behind a state-approved access sheet. It is now, technically, expiring. OpenAI said it plans a broad rollout, has not fixed a general availability date, and expects to open access within days.
Three model tiers ship in the GPT-5.6 family. Sol is the flagship, positioned for coding, biology, and cybersecurity work, with a max reasoning effort mode that the earlier preview docs implied is the actual bar-clearing capability that made the government want to see it first. Terra is the mid-tier option for enterprise workloads where the CFO cares about the price per million tokens. Luna is the fastest and cheapest, aimed at high-volume production tasks where latency matters more than depth. The naming is a slight departure from the numeric-only branding of prior generations, which is one way to acknowledge that the family now behaves differently enough at different price points that customers need proper names for them.
The interesting part of Tuesday was not the clearance. It was that within a few hours of the Axios story, the White House told CNBC that it had given no “green light, approval or clearance” and that model release decisions “rest entirely with the companies.” Both things can be true. Commerce reviewed the model. The White House does not want the reputational weight of appearing to have approved it. That gap between what happened technically and what the executive branch is willing to be seen taking credit for is the shape of AI governance in mid-2026. Everyone wants the process. Nobody wants the fingerprints.
OpenAI, for its part, put out a line saying it does not believe pre-release government access processes should become the long-term default, which is a delicate way of saying “please do not do this again.” The Trump administration framework from June 2 that authorized the review was pitched as voluntary and time-limited. A voluntary framework that the biggest lab in the country publicly hopes goes away is not, in the ordinary meaning of the word, voluntary. It is a preview of a licensing regime with the license paperwork not yet printed. The next model in the queue is Anthropic’s, and everyone in the industry is now watching whether the same 30-day review clock starts running on it.