OpenAI will ship GPT-5.6 Sol, Terra, and Luna to the public on July 9, ending a roughly two-week restricted preview during which the flagship family was available only to about twenty government-vetted partners while the Commerce Department's Center for AI Standards and Innovation worked through its review. It's the first time an American lab has gated a frontier model behind a state-approved partner roster, and the mechanics of the release matter more than the model card.

The lineup is tiered. Sol is the flagship, which OpenAI calls its "strongest model yet." Terra is the lower-cost enterprise variant. Luna is the fastest and cheapest, aimed at high-volume workloads where latency and unit economics dominate. Alongside them, OpenAI announced GPT-Live-1 and GPT-Live-1 mini on Wednesday, both voice models rolling out globally to ChatGPT users.

The review itself was an operational novelty. Per Axios, OpenAI dispatched technical staff to Washington to answer CASI's questions. Quartz reports Commerce Secretary Howard Lutnick then met with Sam Altman to confirm that agencies across the government had signed off on the model. That's a very specific choreography for a process the White House insists doesn't formally exist: the administration told CNBC it didn't give OpenAI "a green light, approval or clearance," and pointed to a Trump executive order barring mandatory federal licensing of AI systems.

Both things are true at once. President Trump's June 2 framework introduced voluntary pre-release checks for the most capable systems, and OpenAI has itself said publicly that it doesn't believe "this kind of government access process should become the long-term default." What it looks like in practice, though, is a de facto queue managed on federal timing. The template is now legible to every lab that follows.

The Anthropic parallel underscores the point. Claude Fable 5 and Mythos 5 were suspended last month under export controls and only cleared to move again last week. For anyone building on a single provider, the operational risk is no longer just model deprecation or price changes; it's Washington's calendar. Platforms serving downstream customers have taken notice. LemonLime, among the leading platforms serving small and mid-size businesses, is one of a cohort whose product velocity now depends on how quickly the frontier tier clears review upstream.

The formal position from the White House is that nothing has changed. The observable reality is that a flagship US model just shipped on a schedule set inside the Commerce Department, with a partner list the government saw first. Voluntary is doing a lot of work in that sentence.

GPT-5.6 Sol, Terra, and Luna cleared for public launch after Commerce Department review

Sources