The three GPT 5.6 tiers — Sol as flagship, Terra as the everyday workhorse, Luna as the cheapest — open to a small circle of government cleared customers, with broader access to follow. OpenAI itself writes that the customer by customer approval process should not become routine.
OpenAI launched its next-generation GPT-5.6 model family on Thursday, but the more important story is the one the company itself flagged in its own announcement: a U.S. government office asked OpenAI to stagger the release, and OpenAI complied before the model reached the public. The company then put a caveat in writing. This kind of customer-by-customer government approval should not become routine.
GPT-5.6 ships in three tiers. Sol is the new flagship. Terra is positioned as the everyday workhorse, with OpenAI claiming competitive performance against the prior generation at roughly half the cost. Luna is the cheapest option in the lineup (OpenAI).
What makes this launch unusual is not the product lineup but the access regime. OpenAI says it previewed plans and capabilities to the U.S. government ahead of today's launch and that broader availability will follow "in the coming weeks." The initial preview is limited to "a small group of trusted partners whose participation has been shared with the government," per the company (OpenAI).
The push for the staggered release came from the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP), according to The Information, which reviewed an internal memo from Sam Altman describing the access as approved "customer by customer" (The Information).
Axios reported that Altman spoke with Commerce Secretary Howard Lutnick on Wednesday, and that the administration asked OpenAI to limit the initial release to government-approved partners (Axios).
Engadget, CybersecurityNews, The Decoder and other outlets corroborated that ONCD and OSTP drove the phased-release request, and that Lutnick advised against launching without cross-agency sign-off (Engadget) (CybersecurityNews) (The Decoder).
OpenAI frames Sol as launching with the company's "most robust safety stack to date," with hardened protections for higher-risk activity, sensitive cyber requests and repeated misuse. That is OpenAI's own framing, not independent validation. The company also describes weeks of internal pressure-testing ahead of the rollout (OpenAI).
The unresolved question is whether the preview window sets a precedent. OpenAI's own post includes the caveat that customer-by-customer approval "should not become the long-term default," which is a notable thing for a company to write inside its own product announcement. The Information's account of the Altman memo frames it the same way, as a temporary measure rather than a new gate (The Information).
The policy environment is not standing still. The release lands as the Trump administration is preparing what OpenAI's post describes as a forthcoming cybersecurity Executive Order framework, and as ONCD and OSTP have already shown they will weigh in on a frontier lab's release schedule. If the next model launch goes through the same doorway, "temporary" will start to look like a category (OpenAI).
Identity of the "trusted partners" beyond the government's own agencies is not disclosed in the available reporting. Benchmarks and exact pricing for Terra and Luna beyond OpenAI's qualitative claims are not supported by the sources in this packet.
For now, GPT-5.6 Sol, Terra and Luna are real and shipping, but only into a small, government-cleared circle. The wider public release in the coming weeks will test whether the preview window was an exceptional precaution or the first draft of a new release norm.