Three-tier release under government oversight
OpenAI rolled out GPT-5.6 in three variants on Friday: Sol, the most powerful version; Terra, balancing efficiency and capability; and Luna, optimized for speed and affordability. All three are subject to access restrictions at the request of the U.S. government, which approved only about 20 companies for the initial preview. OpenAI said it expects to expand access to more organizations in the coming weeks.
Washington's new AI review framework
The restrictions follow a pattern set by the Trump administration's treatment of Anthropic's advanced models. The government has announced plans to evaluate new AI model releases but has yet to detail how the process will work permanently. OpenAI said it has been previewing GPT-5.6 with the government for the past month, including meetings between CEO Sam Altman and the White House in early June. By August, the administration must establish a classified process to assess AI cyber capabilities under a new executive order.
Cybersecurity capabilities spark concern
OpenAI acknowledged that concerns over GPT-5.6's cybersecurity capabilities drove the restrictions, but said the model is better at helping find and fix vulnerabilities than carrying out end-to-end attacks. The company stated that the model's capabilities do not reach the "critical" level outlined in its preparedness framework, and that it expects "substantial benefit for legitimate defensive work" from the release.