OpenAI has unveiled its latest and most advanced model, GPT-5.6, but it is currently not available for access.

      GPT-5.6 introduces enhanced reasoning, autonomy, and cybersecurity features, but its availability is currently restricted to government-approved clients.

      OpenAI has officially launched GPT-5.6, its most advanced AI model family to date. However, there’s a limitation: only a select few approved customers will have access for now. Instead of a wide release, the company is starting with a closely monitored preview while it navigates a new review process mandated by the U.S. government.

      GPT-5.6 is here, but only a limited audience can utilize it

      The GPT-5.6 lineup includes three models: Sol, the premier model designed for the most intensive tasks, Terra for balanced reasoning and routine duties, and Luna, a quicker and more economical option. OpenAI states that GPT-5.6 showcases enhancements in areas such as coding, scientific reasoning, cybersecurity, biology, and prolonged autonomous operations. The flagship Sol model also introduces advanced operational modes: Max for deeper reasoning and Ultra for coordinating sub-agents in intricate workflows.

      However, the most significant aspect isn’t the technology itself, but rather who is permitted access. According to The Wall Street Journal, GPT-5.6 will first be accessible exclusively to a small group of customers vetted by the Trump administration while undergoing additional national security assessments. OpenAI notes that this is a temporary measure during the implementation of new federal oversight and anticipates a broader release of GPT-5.6 in the upcoming weeks.

      In addition to government oversight, OpenAI appears to be taking a stronger stance on security from a technical perspective. Along with the GPT-5.6 Sol model, the company claims it has introduced its "most robust safety stack yet," which enhances real-time safeguards against high-risk cyber activities and attempts at misuse. OpenAI asserts that the model underwent extensive human red-teaming and over 700,000 A100 GPU-equivalent hours of automated safety testing prior to its release.

      The model is remarkable, but the rollout is likely the more significant story.

      The choice to limit access to GPT-5.6, restricting its use to a small group of approved customers, isn’t particularly surprising. Recently, the U.S. government required Anthropic to limit access to its Claude Mythos 5 and Fable 5 frontier AI models due to national security issues. While Mythos has since been reintroduced to selected users, Fable 5 remains unavailable to the general public, restricted to approved entities in the U.S. OpenAI is now pursuing a similar approach.

      "As part of our ongoing discussions with the U.S. government, we previewed our plans and the models’ capabilities before today’s launch. At their request, we’re commencing with a limited preview for a select group of trusted partners whose involvement has been communicated to the government, prior to broader distribution," OpenAI stated in its announcement.

      The company mentioned it will continue to navigate the necessary security vetting process before expanding access to GPT-5.6, although a timeline for a wider release has not been provided. Simultaneously, OpenAI has made it clear that it does not envision this government approval process becoming a standard practice for the release of frontier AI models.

      Besides government scrutiny, OpenAI has another motivation to proceed cautiously. Earlier this week, Anthropic alleged that the Chinese tech giant Alibaba utilized thousands of user accounts to systematically access Claude, distilling its responses to enhance the Qwen AI models. Similar claims have emerged in the past, highlighting growing concerns that frontier AI systems could be copied or misused before their developers can adequately safeguard them. Whether this is a direct influence on OpenAI’s careful rollout remains uncertain, but it is increasingly evident that launching the most advanced AI models goes beyond a mere technical challenge; it is swiftly becoming a geopolitical issue.