OpenAI opened its three-tier GPT-5.6 family to general availability on July 9 across ChatGPT, Codex and its API, ending a 13-day preview window during which access to Sol, Terra and Luna was limited to roughly 20 organisations whose participation had been disclosed to the U.S. government. The company confirmed the wider rollout on X, saying it was “expanding preview access globally now.”

The gating wasn’t a product delay. It was a policy artifact. An AI cybersecurity executive order signed by President Donald J. Trump on June 2 asks companies to voluntarily submit their most powerful models for federal review 30 days before public release. GPT-5.6 became the first flagship-class family to move through that pipeline. The Department of Commerce’s Center for AI Standards and Innovation ran additional testing and met with OpenAI technical staff in Washington before clearing the broader launch, a workflow that reads less like ad-hoc consultation and more like the early scaffolding of a permanent pre-release regime.

OpenAI’s own posture is telling. In a June 26 blog post announcing the restricted preview, the company said it didn’t believe “this kind of government access process should become the long-term default,” framing the arrangement as a short-term step while it works with the administration on a repeatable framework. The message is legible to Washington and to the developer base at once: cooperate now, negotiate the shape later.

Pricing runs $5 and $30 per million input and output tokens for Sol, matching GPT-5.5 at the top; $2.50 and $15 for Terra; $1 and $6 for Luna. A Cerebras-hosted deployment at up to 750 tokens per second is expected later this month. All three tiers are classified “High” risk for cyber and biological/chemical capability in OpenAI’s system card, which is precisely why the review existed, and precisely why the next one will be harder to call voluntary.

Sources