OpenAI receives US approval for an extensive rollout of GPT-5.6 following several weeks of government evaluation.

      OpenAI has received approval to widely release its most advanced model after the US government authorized a broader rollout of GPT-5.6, which had been delayed for weeks due to Washington's new regulations governing frontier AI. Until now, the model was only accessible through a limited preview to approximately 20 partners whose identities were individually authorized by the US government. This arrangement, a first for an American frontier model, is now being replaced by the broader release.

      The approval came after further testing conducted by the Commerce Department’s Center for AI Standards and Innovation, the agency responsible for assessing advanced systems. OpenAI dispatched technical experts to Washington to address the agency’s inquiries, as reported by Axios.

      GPT-5.6 consists of a three-tier family rather than one singular model. The flagship is Sol, Terra serves as a lower-cost mid-tier option, and Luna is the fastest and least expensive of the trio.

      OpenAI has highlighted Sol's strengths in coding, biology, and cybersecurity, and has introduced a "max reasoning effort" mode that allocates more time for solving difficult problems. These capabilities, especially in biology and cybersecurity, are part of why the government sought a closer examination prior to the wider release.

      The tier system reflects both commercial and technical choices. Terra targets typical enterprise workloads where cost is more critical than raw power, while Luna is designed for high-volume tasks that prioritize speed, allowing OpenAI to offer varied pricing within the same family.

      The previous preview was notably stringent. For several weeks, GPT-5.6 was exclusively available to a select few organizations whose names OpenAI disclosed to the government, marking the first instance of an American lab restricting a frontier model based on a state-approved list.

      This review process was established within a framework set by the Trump administration on June 2, which introduced a voluntary pre-release check for the most advanced models. The case of GPT-5.6 exceeded this by transitioning from a voluntary review to a government-managed access list, a step OpenAI agreed to after being asked to postpone the launch.

      OpenAI has indicated it is uncomfortable with this precedent, expressing that it does not believe such a government access process should become a long-term norm, despite its agreement to participate this time.

      The unease is understandable; a government that can regulate a launch can also halt one, a power already exercised by the administration in another scenario by instructing Anthropic to discontinue two models.

      For OpenAI, the financial implications of the delay were substantial. Each week GPT-5.6 remained in a limited 20-partner preview was an opportunity for competitors to attract the enterprise customers OpenAI aimed to engage with its new tiers.

      The company now plans to expand access to GPT-5.6 within days, building on the groundwork laid with GPT-5.5 earlier this year. It has stated that all three tiers will become generally accessible in the coming weeks, although no specific public date has been announced.

      OpenAI is not the sole lab operating under the new regime; this framework also applies to its competitors, influencing how future frontier models from any US company will reach the public.

      This situation establishes less of a product timeline and more of a template. For the first time, a prominent US lab has released a frontier model according to the government’s timeline rather than its own, raising questions about whether this approach will be a one-time occurrence or indicative of future trends.