OpenAI opened a limited preview of GPT-5.6 on Friday, releasing its Sol, Terra and Luna models only to a small group of vetted partners after the Office of the National Cyber Director and the Office of Science and Technology Policy asked the company to gate access on national-security grounds. It’s the first time a U.S. frontier-model launch has been throttled by the federal government before general release.

The company registered its discomfort in the same blog post that announced the preview. “We don’t believe this kind of government access process should become the long-term default,” OpenAI wrote, warning that the arrangement “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.” Sam Altman, in an internal memo cited by Reuters and The Information, told staff that access would be approved “customer by customer during this preview period.”

The gating follows an export-control directive earlier this month that forced Anthropic to disable Mythos and Fable. A source told CNN the administration views Sol as “on par” with Mythos. An executive order President Trump signed earlier in June asks developers to voluntarily submit advanced models for review 30 days before release, but the implementing framework doesn’t yet exist. The preview is effectively that framework being built in public, customer by customer.

OpenAI says Sol doesn’t reach the “critical” threshold in its preparedness framework and expects general availability in the coming weeks, at $5 per million input tokens and $30 per million output tokens. The partners weren’t disclosed. The company is calling the gating a short-term step toward a “repeatable process for future model releases,” which is the part worth watching: repeatable processes, once built, tend to outlive the emergencies that justified them.

Sources