OpenAI's New GPT-5.6 Models Launch Behind a Government Access List

OpenAI has begun previewing its next generation of models, GPT-5.6, but with a twist that says as much about 2026 as it does about the technology: for now, you can only use it if the US government approved you by name.

The company unveiled three variants, Sol, Terra and Luna, and handed early access to roughly 20 partners whose participation was individually cleared by Washington. It is, as several outlets noted, the first time an American AI company has launched a frontier model behind a government-managed access list.

Model	Built for
Sol	The heavyweight, aimed at the hardest problems, complex coding and security research
Terra	High-volume business work such as customer support, internal tools and document analysis
Luna	The quick, cheaper option for everyday jobs like summarizing, drafting and routine automation

What are the three GPT-5.6 models?

OpenAI has split the family by task. Sol is the heavyweight, aimed at the hardest problems, complex coding and security research. Terra is built for high-volume business work such as customer support, internal tools and document analysis. Luna is the quick, cheaper option for everyday jobs like summarizing, drafting and routine automation.

During the preview the three are reachable only through OpenAI's API and its Codex coding agent, and only for that short list of trusted partners and organizations. General availability, the company says, is coming "in the coming weeks."

Why is GPT-5.6 limited to government-approved partners?

The restrictions grew out of OpenAI's own engagement with federal officials. The company said it previewed the models' capabilities to the government ahead of launch, and Sol in particular, with its strength in coding and cybersecurity, is the kind of tool authorities worry about in the wrong hands.

OpenAI has been careful to frame the arrangement as an exception rather than a template. Access limits like these, it said, should not become the norm. That is a notable thing for a leading lab to feel it must say out loud, and it hints at an uncomfortable new reality: the most capable systems are now powerful enough that their rollout is a matter of state interest, not just product strategy.

What it signals

Strip away the model names and a pattern emerges across the industry this summer. Anthropic briefly went dark over export controls; OpenAI is now shipping its best models through a government-vetted list. The frontier labs and national governments are becoming entangled in a way they were not even a year ago.

For everyone else, the practical takeaway is a waiting game. GPT-5.6 is real, it is being used, and the benchmarks and demos will trickle out through the approved partners. But the version most people can actually touch is still a few weeks, and one bureaucratic green light, away.