OpenAI has released GPT-5.6, which it calls its strongest model family yet, but the rollout is deliberately restricted, according to multiple reports.
The new lineup includes three tiers: Sol, the flagship; Terra, a lower-cost option; and Luna, the fastest. Instead of a wide public launch, OpenAI is taking a staged approach. According to Yahoo, the release is restricted to the United States only. According to CBC and other outlets, access is being limited to a Trump-approved group of customers during a testing period tied to a cybersecurity review.
Reports carried by WWMT, WPEC and The Hill say OpenAI is slow-rolling the model at the Trump administration's request. The Star Beacon and Texomashomepage.com describe the limited access as covering Trump-approved customers during a cybersecurity review.
The security concern centers on the models' capabilities. According to OpenAI (via Techmeme), GPT-5.6 Sol and Terra were capable of identifying vulnerabilities but were unable to execute autonomous, end-to-end attacks against hardened targets. In other words, the models can spot weaknesses but, per the company, could not independently carry out full cyberattacks.
OpenAI says a wider release is expected in the coming weeks, according to MSN, which framed the staged launch against the backdrop of an escalating AI arms race. Nextgov/FCW reports the model is being released to select partners.
Why it matters: This is an unusual case of a frontier AI model being gated by government review before broad release, signaling that cybersecurity risks and federal oversight are starting to shape how and when the most capable AI tools reach the public.