OpenAI unveils GPT 5.6 with Sol, Terra and Luna models; broad rollout delayed after US request

OpenAI has introduced a new family of artificial intelligence models led by GPT 5.6, comprising Sol, Terra and Luna, while limiting initial access to a small group of trusted partners following a request from the US government.The company said the restricted rollout is a temporary measure and that wider availability is expected in the coming weeks. According to OpenAI, the US government requested a limited preview because of concerns around the advanced capabilities of the new models.In a statement, the company said it had shared its launch plans and model capabilities with the US government ahead of the release. "At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly," OpenAI said, adding that it would continue testing the models while coordinating with partners during the preview period.OpenAI also reiterated that it does not believe government-controlled access should become the long-term standard for releasing advanced AI models."We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them," the company said, adding that it is working with the US administration to establish a repeatable framework for future frontier AI releases.The newly announced model family is led by Sol, which OpenAI described as its most capable AI model to date. The company said Sol delivers strong performance across areas including software engineering, biology and cybersecurity, and claimed it outperformed Anthropic's Mythos model on the Terminal Bench 2.1 benchmark used to evaluate complex AI tasks.Also read: Anthropic restores Claude Mythos 5 access after the US government eases restrictionsOpenAI also introduced Terra, positioned as a balanced model for everyday workloads, and Luna, designed to offer faster performance at a lower cost. According to the company, both models incorporate its latest safety protections, with safeguards tailored to each model's capabilities.OpenAI said it continues to strengthen security measures as its AI systems become more powerful, while ensuring the models remain useful for legitimate applications such as software development, cybersecurity testing, code reviews, vulnerability identification and system maintenance.