OpenAI has begun previewing a new model series called GPT-5.6 in three variants: Sol, Terra, and Luna. According to the company, Sol is its most capable model to date; Terra is positioned for everyday use at roughly half the price of GPT-5.5; and Luna is the lowest-cost option.
The preview was limited. OpenAI said it showed the models to a "small group of trusted partners," and the first party to see the new models was the United States government. At the administration’s request, OpenAI shared the partner list with Washington prior to the public release. The company described this as a short-term step it does not want to become the long-term default, framed as a way to make the new models available to the public sooner.
Major labs provided early access ahead of a voluntary order
The New York Times reports that multiple major AI labs — including Anthropic, Google, xAI, and Microsoft — were already granting early access before a federal cybersecurity "voluntary" order was signed. Meta was reported to be the lone holdout, and the government is said to be pressing it to comply.
When most leading labs follow a practice that is not legally binding, the label "voluntary" loses practical meaning: the gate for frontier AI releases is quietly shifting from corporate decision-making to federal review.
Why this matters
- There has been no legislative vote or public statute determining how the most powerful models should be released; the current practice emerged without formal public deliberation.
- The fact that major developers are giving Washington early access suggests that federal review is becoming a de facto step before deployment of advanced models.
- The longer-term implications — including effects on access, transparency, and competition — remain unresolved.
The episode highlights that the release pathway for the most capable AI models increasingly involves government review, even as formal legal frameworks and broad public consensus are still absent.
