OpenAI begins limited preview of GPT-5.6 models with U.S. government partners

OpenAI announced a limited preview of its GPT-5.6 model series to select U.S. government partners, framing the move as part of an ongoing engagement with federal officials. The company described the initiative as a short-term step to coordinate with collaborators before broader availability, which it expects in the coming weeks.

The preview includes three models—Sol, Terra, and Luna—each tailored for different use cases. OpenAI describes Sol as its "strongest model yet," with strong performance in coding, biological research, and cybersecurity tasks. Terra is positioned as a lower-cost alternative with coding and output token performance close to Sol, while Luna is characterized as the fast and most cost-efficient option in the series.

OpenAI states that each model includes safeguards designed to resist adversarial pressure while preserving access for legitimate activities such as code review, vulnerability research, patch development, debugging, security education, and defensive testing.

The announcement follows a White House executive order signed earlier this month that calls on leading AI developers to share model access before market release for safety analyses. OpenAI framed its limited preview as aligned with this directive, emphasizing that it does not intend for restricted government access to become a long-term default.

OpenAI’s move comes amid heightened federal scrutiny of AI developers, including the designation of Anthropic as a supply chain risk and the imposition of export controls on some of its advanced products. The company’s Sol model and a pending "Sol Ultra" variant are positioned as responses to Anthropic’s Mythos model, which prompted the administration’s recalibration of its relationship with the company.