The approach is similar to Anthropic’s Project Glasswing and follows a Trump administration request that the company limit access. #
OpenAI is offering a limited preview of its new GPT-5.6 model series to select partners, part of an “ongoing engagement with the U.S. government” as Washington and leading artificial intelligence developers try to strike a balance between tech innovation and safety.
Three models in the company’s GPT-5.6 series — Sol, Terra and Luna — will initially be available to select partners following conversations and previews with government officials, OpenAI said in its Friday announcement.
That initially limited rollout follows a request made by the federal government. During the narrow preview period, OpenAI will test model capabilities and coordinate with collaborating partners before making it more broadly available. General access to the models will be available “in the coming weeks,” a similar structure to Anthropic’s Project Glasswing that was set up to test its Mythos Preview model.
OpenAI made it clear that it primarily believes in broad, open access to AI models and that limited previews for government partners are short-term.
“We don’t believe this kind of government access process should become the long-term default,” OpenAI said in its announcement. “It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.”
The three GPT-5.6 series models are each tailored for different uses. The flagship GPT-5.6 Sol is OpenAI’s “strongest model yet,” and performs well in coding, biological and cybersecurity work, according to the company.
Terra serves as a lower-cost option whose performance in coding workflows and output tokens falls closely behind that of Sol, according to data published by OpenAI. The final model, Luna, is the fast and most cost-efficient model in the series.
Each of the models features robust safeguards, OpenAI says, designed specifically to resist “real-world adversarial pressure while preserving access to legitimate work such as code review, vulnerability research, patch development, debugging, security education, and defensive testing.”
OpenAI created Sol — and a pending version, Sol Ultra — to rival Anthropic’s landmark Mythos model, a powerful AI system that prompted the Trump administration to recalibrate its fraught relationship with Anthropic, after the government designated the company a supply chain risk and imposed export controls on several of its more advanced products.
OpenAI’s announcement follows Trump's signing of an executive order earlier this month that calls for leading AI developers to share model access before market release for safety analyses. Signed following last-minute industry pushback regarding overregulation concerns, the order and OpenAI’s choice to voluntarily share its new model with the government signal the priority the White House is placing on AI safety despite trying to thread the needle between a light-tough regulatory posture and ensuring companies are unencumbered enough to innovate.