OpenAI announced three new artificial intelligence models on Friday — GPT-5.6 Sol, Terra, and Luna — but said it is limiting initial access to a small group of government-approved "trusted partners" at the Trump administration's request.
The company plans to make GPT-5.6 Sol, Terra, and Luna broadly available in the coming weeks, but says the vetting process shouldn't become the long-term default
Cheng Xin / Getty Images
OpenAI announced three new artificial intelligence models on Friday — GPT-5.6 Sol, Terra, and Luna — but said it is limiting initial access to a small group of government-approved "trusted partners" at the Trump administration's request.
The company said it previewed the models' capabilities with government officials ahead of Friday's launch and agreed to a phased rollout as it works toward broader public availability in the coming weeks. OpenAI did not disclose the names of the approved partners. But access is being extended to 20 partners, with Amazon $AMZN's Bedrock software platform serving as one route in.
"We don't believe this kind of government access process should become the long-term default," OpenAI said. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
Among the three models, Sol sits at the top of the capability hierarchy, while Terra occupies a middle tier and Luna is designed for lower-cost use cases. On cybersecurity, OpenAI described Sol as more useful for detecting and patching software flaws than for executing attacks from start to finish, with additional capability gains also noted in coding and biology. The company said none of the three models cross its internal "critical" cybersecurity risk threshold.
Even so, OpenAI placed all three models in the high-capability category for cybersecurity and biological and chemical risk under its Preparedness Framework, while noting that novel dangers could surface — especially when the models are used alongside other technologies.
OpenAI CEO Sam Altman and Commerce Secretary Howard Lutnick discussed the GPT-5.6 launch on Wednesday, according to The Wall Street Journal, which also reported that the government's Center for AI Standards and Innovation had the models under review.
The staggered release follows a turbulent stretch for the AI industry stemming from the government's handling of Anthropic's models. Earlier this month, the Trump administration barred foreign nationals from accessing Anthropic's Mythos 5 and Fable 5, citing concerns about the models' potential to be exploited for malicious cyberattacks. Enforcing a nationality-based restriction turned out to be technically impossible, leaving Anthropic with no option but to pull both models from service entirely — all after receiving only about 90 minutes of warning.
An executive order Trump signed earlier in June created a process under which AI companies could voluntarily offer their most powerful models to federal reviewers for as long as 30 days prior to any public launch — though the Journal reported that the rules underpinning that process have yet to be finalized. The company said its near-term goal is to establish, alongside the administration, a consistent set of procedures that can govern how future models are assessed and released.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.