OpenAI launches GPT-5.6 after U.S. security review

OpenAI will publicly release GPT-5.6, its most advanced AI model family, on Thursday after the Trump administration cleared the rollout following weeks of security testing and consultations between the company and government officials. The launch marks one of the first times a major AI model has undergone a government-led review before being made widely available to the public.reuters+2

A Delayed Release

The GPT-5.6 series — comprising the flagship Sol model, the mid-tier Terra, and the cost-efficient Luna — was initially announced in late June but restricted to a small group of vetted partners at Washington's request. National security officials raised concerns that the model's advanced capabilities, particularly in cybersecurity and reasoning, could be exploited for cyberattacks or misused by foreign military and intelligence agencies.cybernews+3

The Commerce Department approved the broader launch after additional testing under a framework stemming from a June 2 executive order by President Donald Trump requiring the Defense Department to design a system for AI developers to voluntarily share access to frontier models with the government before public release. Government officials were given 30 days to review the technology and flag potential risks.cnet+1

"We're expanding preview access globally now," OpenAI said in a post on X announcing the Thursday launch. CEO Sam Altman confirmed the release date on Wednesday.engadget+1

What the Models Offer

GPT-5.6 Sol is OpenAI's most capable model, designed for complex tasks including advanced coding, research, and long reasoning chains. Terra offers performance comparable to GPT-5.5 at roughly half the cost, aimed at everyday work like writing and document analysis. Luna is the fastest and cheapest option, built for quick conversations and routine tasks.economictimes+1

OpenAI described the series as a "meaningful step up in cybersecurity capabilities," with strengthened safeguards to prevent misuse while assisting defensive security work. The company also said Sol produces fewer factual errors compared to previous models.cnet+1

A New Precedent

The delayed rollout sets a precedent for how Washington may handle future AI releases. According to CNET, this is the first time the US government has held back a company with requests for review before a model's wide release. The process has drawn attention to the tension between rapid AI development and national security oversight — a dynamic likely to intensify as models grow more powerful.cnet