OpenAI has released GPT-5.6, its most capable model family to date -- comprising a flagship model called Sol, a balanced everyday model called Terra, and a faster, lower-cost model called Luna -- but in an unprecedented move, the initial rollout is limited to a small group of trusted partners whose participation is disclosed to the US government. The restriction came at the request of the Trump administration, and OpenAI complied while making clear it did so reluctantly.
The company did not hide its discomfort. In its announcement, OpenAI said it does not believe 'this kind of government access process should become the long-term default,' framing the limitation as a temporary measure adopted while it works with the administration on cybersecurity frameworks for advanced models. Broader availability across ChatGPT, the Codex coding tool, and the API is promised 'in the coming weeks' -- but for now, the most powerful version of the technology sits behind a federal gate.
The pricing makes clear these are serious frontier systems aimed at developers and enterprises. Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra runs at roughly half that; and Luna, the budget tier, costs $1 and $6 respectively. Those are premium rates that signal capability, but the headline isn't the price -- it's that, for the first time, a buyer's ability to access a leading American model depends on clearing a government-administered list.
“Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra runs at roughly half that; and Luna, the budget tier, costs $1 and $6 respectively.”
This did not happen in isolation. The same week, Commerce Secretary Howard Lutnick authorized the release of Anthropic's Mythos 5 cybersecurity model to more than 100 US companies and agencies, two weeks after the administration had banned it outright. Both episodes trace to the same anxiety: that the cyber-offense capabilities of frontier models are now potent enough that Washington wants to control who wields them. OpenAI's GPT-5.6 restriction is the second shoe to drop, and it suggests a pattern rather than a one-off.
The critique was immediate and came from inside the tent. Dean Ball, a former White House AI adviser and incoming OpenAI employee, warned that the approach risks creating a 'de facto involuntary licensing regime' for AI -- a world in which the government, not the market, decides who may deploy the most capable systems. That is a profound shift from the open, developer-first distribution model that built the modern AI industry, where anyone with a credit card could call the best model through an API.
For founders and investors, the implications are sharp. A generation of startups was built on the assumption of frictionless access to frontier models. If access becomes a privilege extended to vetted partners, the competitive landscape tilts toward incumbents and government-aligned firms, and foreign developers may be cut off from the best American tools entirely -- a potential boon to Chinese labs like Alibaba's Qwen and DeepSeek that ship open weights. The moat stops being purely technical and becomes partly political.
The bear case for over-reading it: OpenAI calls this temporary, broader rollout is weeks away, and a cybersecurity carve-out for the single most capable tier is not the same as licensing all AI. Governments have long restricted dual-use technologies, from encryption to satellite imagery, without freezing the underlying industry. The question is whether 'temporary' holds, or whether each successive frontier release now ships through the same national-security filter by default.
What to watch: whether GPT-5.6's broad availability actually arrives on schedule and on the same terms for everyone, how the administration defines and polices its 'trusted partner' list, and whether the next frontier model from any lab launches free or gated. If gating becomes the norm, the most important variable in AI stops being model quality and becomes permission -- and that rewrites the rules for every company building on top of these systems.