OpenAI Retires GPT-4.5, Launches Vetted GPT-5.6 Under US Oversight

Model Sunset and a Regulated Dawn

OpenAI is initiating a significant model transition, confirming the retirement of its GPT-4.5 model on June 27, 2026, and the o3 model on August 26, 2026. This planned sunset coincides with the highly unusual, government-influenced rollout of its successor generation.

In a move echoing the launch of Anthropic's Mythos, OpenAI is deploying its new GPT-5.6 model suite under a strict "limited preview." This staggered release comes at the direct request of the Trump administration, as confirmed by CEO Sam Altman in an internal memo obtained by The Information.

A Trio of New Models Under Scrutiny

The GPT-5.6 suite, unveiled less than 24 hours after news of the regulatory request broke, consists of three distinct models designed for different use cases. The flagship model is Sol, engineered for high-performance tasks. Terra is positioned as a medium-tier model optimized for "high-volume work," while Luna is marketed as a "fast and affordable" option for everyday applications.

OpenAI highlights that the new suite is "especially skilled at coding, cybersecurity, and biology," as well as maintaining focus during long-horizon, agentic AI tasks. This advanced capability in cybersecurity, in particular, is central to the government's heightened scrutiny.

The 'Case-by-Case' Government Vetting Process

The core of this unprecedented launch is the federal oversight. According to Altman's memo and subsequent company statements, the US government will be "approving access customer by customer during this preview period." OpenAI is offering the preview to "a small group of trusted partners whose participation has been shared with the government."

This process means potential users are being vetted on a case-by-case basis by the Trump administration before gaining access to the powerful new models. The company stated it "cooperated with the US government ahead of this launch" but emphasized this is "not our preferred long term model."

continue reading below...

Balancing Defense and Offense in AI Security

OpenAI's official rationale focuses on a delicate balance. The company asserts that GPT-5.6 Sol is "better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks." Their stated priority is to ensure these advanced capabilities "reach and benefit defenders" who can use them to patch weaknesses and strengthen systems.

However, the company acknowledges the dual-use dilemma. In a statement, OpenAI wrote that "safeguards may occasionally intervene on legitimate work, particularly in dual-use areas where defensive and offensive activity can initially look similar." They state their goal is to "make prohibited offensive activity more difficult, uncertain, and detectable without unnecessarily limiting those beneficial uses."

Industry Implications and Future Release Strategy

This event marks a pivotal moment in AI governance, setting a potential precedent for future frontier model releases. OpenAI has indicated that if the vetting process proceeds smoothly, a broader, general availability release for GPT-5.6 is planned for "the coming weeks" or "a couple of weeks later," as the company believes in "broad access."

The staggered, government-vetted release represents a significant shift from OpenAI's previous launch strategies. It underscores growing regulatory concerns about the national security implications of advanced AI, particularly models with potent cybersecurity capabilities. How this approach affects the competitive landscape and the pace of AI innovation remains a key question for the industry.

A New Chapter in AI Regulation

The retirement of GPT-4.5 and o3 symbolizes the end of one era, while the controlled launch of GPT-5.6 under government supervision may signal the beginning of another. This development illustrates the increasing entanglement of cutting-edge AI development with national security policy.

For developers and enterprises, it introduces a new layer of complexity in accessing state-of-the-art tools. For policymakers, it represents a tangible implementation of AI oversight. The coming weeks, as the preview unfolds and the general release decision is made, will be critical in shaping the future relationship between AI labs and government regulators.