OpenAI has limited the initial release of three new artificial intelligence models to a small number of "trusted partners". On June 26, CNBC reported that OpenAI, at the request of the U.S. government, narrowed the early rollout of GPT-5.6 Sol, Terra and Luna and said it would pursue a public release in the coming weeks.

OpenAI said in a blog post that it shared the models' performance and launch plans with the government in advance of the release. It added that such government access procedures should not become the long-term norm, saying they could prevent users and developers, companies, cyber defense organisations and global partners from accessing needed tools. It did not disclose the list of partners allowed early access.

The announcement comes alongside Anthropic's recent steps. Anthropic said two weeks ago it halted access to two of its latest models to comply with Trump administration export control guidelines. Anthropic is in talks with officials in Washington, D.C., but did not say when service for the models would resume.

The Trump administration has been intervening more actively in AI regulation after U.S. President Donald Trump signed an AI executive order earlier this month. The new executive order also includes a request for voluntary cooperation from developers so the government can assess model capabilities before formal public release. OpenAI said it is setting up an evaluation framework with the Trump administration and developing procedures that can be applied repeatedly for future model launches.

Of the three models OpenAI is introducing, Sol is the most powerful product. OpenAI said Sol has improved performance in coding and biology, and has the highest capability among its models in cybersecurity.

OpenAI said Sol did not exceed the company's internal threshold for "critical" cybersecurity risk.