OpenAI limits – OpenAI says it is restricting GPT-5.6 to a small group of “trusted partners” at the U.S. government’s request, while the company argues the process shouldn’t become the default for frontier AI releases. GPT-5.6 includes Sol, Terra, and Luna, with preview acces
On Friday. OpenAI said it will hold back the rollout of its newest AI models—starting with GPT-5.6—after a request from the U.S. government. Instead of a broad preview. the company is limiting access to a “small group of trusted partners. ” with those partners’ participation “shared with the government.”.
The next generation lineup includes three models: Sol. described as OpenAI’s flagship; Terra. positioned as a more balanced everyday option; and Luna. pitched as a faster. lower-cost alternative. OpenAI also said Sol is its most powerful model, but the Trump administration restricted the release of all three.
The timing lands amid mounting federal pressure on AI companies to rein in their most advanced systems. The U.S. government also ordered Anthropic to remove access to its most powerful public model. Fable 5. for any foreign national after Anthropic released it. Anthropic responded by taking the model down entirely.
For OpenAI, this moment isn’t just about logistics. The company pushed back directly against the mechanism itself, even while complying. In a Friday blog post. OpenAI wrote: “We don’t believe this kind of government access process should become the long-term default. ” adding that it “keeps the best tools from users. developers. enterprises. cyber defenders. and global partners who need them.”.
Dean Ball. a former White House AI adviser and soon-to-be OpenAI employee. said President Trump’s recent executive order—asking certain AI companies to voluntarily submit their most advanced models for government review up to 30 days before release—has created a “de facto involuntary licensing regime for frontier AI.” Ball’s concern is that the government doesn’t have clearly defined safety standards. which could turn the approval pipeline into endless launch delays. He argued that delays could help China in the AI race and also jeopardize the billions of dollars going to AI infrastructure buildouts.
OpenAI called Friday’s preview restriction a “short-term step. ” saying it plans to put GPT-5.6 on a path to broader availability in the coming weeks. The company also said it is working with the administration to develop a new executive order framework focused on cybersecurity. along with a “repeatable process for future model releases.”.
Inside the GPT-5.6 bundle, OpenAI said Sol brings improved agentic capabilities in coding, biology, and cybersecurity. OpenAI described Sol’s new reasoning features as a “max” reasoning effort mode and an “ultra” mode that uses coordinated subagents to solve highly complex tasks—an approach OpenAI noted could drive token usage higher. The company said Sol performs well across multiple benchmarks. claiming it is slightly better at coding workflows than Anthropic’s Claude Mythos 5. which the Trump administration also effectively banned this month. OpenAI also said GPT-5.6 Sol is competitive with Mythos preview but uses a third of the output tokens.
To address worries about unsafe capabilities. OpenAI said Sol includes its “most robust security stack yet.” It described the system as heavily hardened against adversarial attacks and “intentionally optimized to favor defensive cybersecurity work over offensive exploits.” In OpenAI’s framing. the model is designed to be hard to jailbreak while prioritizing how to defend against exploits rather than teaching users how to hack into systems.
OpenAI further said its safety guardrails are built into the core model’s behavior. not layered on top through a separate filter. The firm pointed to what happened with Fable 5: during the brief moments when it was available. when the model’s classifiers detected a high-risk topic such as cybersecurity. biology. or chemistry. it wouldn’t just block the prompt. OpenAI said Fable 5 would route requests to an older model. and OpenAI argued that the over-cautious flow and invisible downrouting led to false positives and user backlash.
Even with Sol restricted to partners for now, OpenAI said it intends to make the models more broadly available soon for people using ChatGPT, Codex, and the API.
GPT-5.6 is offered in three sizes with tiered pricing. OpenAI said Sol costs $5 per million input tokens and $30 per million output tokens. Terra costs half those amounts, and Luna costs $1 per million input tokens and $6 per million output tokens. OpenAI also said it has improved prompt caching to make repeated prompts cheaper and more predictable.
The debate around who controls advanced AI isn’t theoretical anymore. With GPT-5.6 limited at the start—and with the government already forcing Anthropic to take Fable 5 down—OpenAI is betting that the next version of the rules will be clearer. For now. though. the models that OpenAI says are built for the toughest work will only reach a small. approved circle.
OpenAI GPT-5.6 Sol Terra Luna U.S. government AI review frontier AI cybersecurity Anthropic Fable 5 executive order token pricing prompt caching ChatGPT Codex API jailbreaking agentic capabilities