OpenAI released its GPT-5.6 model on July 9, following a preliminary disclosure to select U.S. institutions on June 26. The release came after direct coordination with Trump administration officials who requested a phased rollout to assess national security risks - a process that signals how government oversight of advanced AI has moved from exceptional to routine.
The model arrives in three configurations: the top-tier Sol, the mid-range Terra, and the cost-efficient Luna. This tiered approach mirrors Anthropic's strategy of offering both unrestricted and partially restricted versions of its next-generation models. OpenAI said Sol's performance matches Anthropic's Mythos 5, the unrestricted counterpart to Fable 5, which the U.S. government recently placed under export controls.
Benchmark results place it near the top
OpenAI's published data shows Sol scoring 88.8% on Terminal-Bench 2.1, a coding ability test, edging past Mythos 5 at 88%. On CyberGym, which evaluates cybersecurity capabilities, Sol hit 84.5% against Mythos 5's 83.8%. Independent evaluator Artificial Analysis ranked GPT-5.6 Sol at 59 points, second only to Fable 5 at 60 points.
Government consultation shaped the release
Under the Trump administration's pre-consultation framework for advanced AI models, OpenAI discussed both the timing and method of release with federal officials. The government asked the company to stagger the launch rather than push a full public release immediately, citing additional verification of national security risks. OpenAI confirmed it "consulted extensively with the U.S. government during the GPT-5.6 release process."
For government professionals tracking how AI governance is evolving, the AI Learning Path for Policy Makers covers the regulatory frameworks now shaping these release decisions. The coordination between OpenAI and multiple federal agencies reflects a working model that other AI firms are likely to encounter.
CEO Sam Altman told CNBC he worked directly with U.S. Secretary of Commerce Howard Lutnick, U.S. Treasury Secretary Scott Bessent, and National Cyber Director Sean Kerncross. "We incorporated many changes, and the government is identifying potential issues," Altman said.
The British daily the Guardian described the process as evidence of "the U.S. government's growing concern over security risks related to AI."
Why this matters for government professionals
The GPT-5.6 release establishes a concrete precedent: frontier AI models now move through government review before reaching the public. For agencies and policymakers, this means pre-consultation frameworks are no longer theoretical. They are operational and shaping product timelines. The tiered release structure - full capability, mid-tier, and cost-efficient - also creates a de facto classification system that procurement teams, security reviewers, and regulators will need to understand when evaluating which model versions can be deployed in government contexts. Resources like AI for Government track how these classification and review processes are developing across agencies.
Your membership also unlocks:
