For now, the company is giving access to only a small group of trusted partners approved by the US government, while extended release is expected in the coming weeks.
The limited launch happened after a request from the Donald Trump administration to temporarily restrict GPT 5.6 because of concerns about its advanced abilities. While OpenAI agreed to the short-term plan, the company also made it clear that it did not believe government approval should become the normal way people gain access to new AI models.
OpenAI said this was only a temporary plan, adding it was working with the US government for a better long-term system for future releases. The company added that keeping its most advanced tools available to only a small group was not the best solution, as developers, businesses, researchers, security experts and other users also need access.
In a statement, OpenAI said, “As part of our ongoing engagement with the US government, we previewed our plans and the models’ capabilities ahead of today’s launch. 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 widely. During this preview, we will continue testing and coordinating closely with partners as we work toward broader availability.”
“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. 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.”
OpenAI said Sol is the most powerful AI model it has created so far. According to the company, it performs well in areas like computer programming, biology and cybersecurity. OpenAI also said Sol scored better than Anthropic’s Mythos model on Terminal Bench 2.1, a test used to measure how well AI systems can handle difficult tasks.
Terra and Luna were also built with their strongest safety measures so far. The company explained that each model has safety settings designed to match its level of capability.
As its AI models become more advanced, OpenAI said it continues to improve these protections so they can better handle misuse and real world attacks.
At the same time, the company said it wants to make sure people can still use the models for legitimate work, such as reviewing code, finding security problems, fixing software bugs, creating updates, learning about cybersecurity and testing systems to improve their security.
