OpenAI shipped GPT-5.6 Friday, then handed the keys to a small group of trusted partners at the U.S. government’s request. The release covers three tiers: Sol, the flagship built for agentic work in coding, biology, and cybersecurity; Terra, which performs similarly to GPT-5.5 at half the cost; and Luna, the lowest-priced option in the lineup. General availability is weeks out.
The restriction flows from Trump’s June 2 executive order on AI and cybersecurity, which created a voluntary review window for the government to assess frontier models. OpenAI says it’s working with the administration to build a “repeatable process” for future model releases. Friday’s arrangement is the first live test of that framework.
OpenAI didn’t hide its frustration: “We don’t believe this kind of government access process should become the long-term default,” the company wrote, noting the cost to developers, cyber defenders, and global partners who need these tools now. It’s cooperating because it believes short-term restriction is the fastest path to broad release.
The Anthropic example shows how far the framework can reach. In June, a US export control directive forced Anthropic to disable Fable 5 and Mythos 5 for all foreign nationals, including its own non-citizen employees, after the government cited a jailbreak in the model’s cybersecurity safeguards. Cooperative turned coercive inside two weeks.
If you’re building on frontier model APIs, don’t plan around general availability dates. Plan around whether your product survives a three-week hold when a jailbreak surfaces in the model powering it.
— Nathan Zakhary