OpenAI Releases GPT-5.6 Model Suite in Limited Preview Under US Government Oversight

OpenAI Launches Limited Preview of GPT-5.6 Model Suite

The restricted release follows a request from the Trump administration and comes less than 24 hours after reports that the company agreed to stagger its deployment.

OpenAI released a limited preview of its GPT-5.6 model suite on Friday, restricting access to a small group of government-approved partners.

Models and Capabilities

The GPT-5.6 suite includes three models:

• Sol (flagship): Designed for advanced reasoning and complex tasks. Includes a "max" mode for deeper reasoning and an "ultra" mode that coordinates sub-agents.
• Terra (mid-tier): A balanced model positioned between Sol and Luna in capability and cost.
• Luna (fast/affordable): Optimized for speed and lower computational cost.

OpenAI stated the models demonstrate improved skills in coding, cybersecurity, biology, and long-horizon agentic tasks.

Safety and Oversight

OpenAI dedicated a significant portion of its announcement to safety measures. The company stated that GPT-5.6 models are trained to refuse prohibited cyber assistance, including jailbreak attempts. According to OpenAI, Sol does not cross the cyber-critical threshold under its revised preparedness framework.

The company reported using approximately 700,000 A100e GPU hours for automated red-teaming and is working with third-party testers over a two-week period.

The preview period is monitored by the Trump administration, which will approve customers on a case-by-case basis. OpenAI stated that safeguards may occasionally intervene on legitimate work in dual-use areas.

Government Context

The administration's request follows recent actions against other AI companies. After Anthropic released its public model Fable 5, the administration ordered the company to remove access for foreign nationals, leading Anthropic to take the model down entirely.

OpenAI stated that GPT-5.6 Sol slightly outperforms Anthropic's Claude Mythos 5 in coding workflows while using one-third of the output tokens.

Dean Ball, a former White House AI adviser who is scheduled to join OpenAI, stated that President Trump's recent executive order—which asks AI companies to voluntarily submit advanced models for government review up to 30 days before release—has created a de facto licensing process. Ball argued that without clearly defined safety standards, this process could cause delays that may benefit China in the AI development race.

OpenAI's Position

OpenAI expressed dissatisfaction with the current arrangement. In a Friday blog post, the company stated:

"We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."

OpenAI described the preview as a "short-term step" toward broader availability in the coming weeks. The company stated it is working with the administration to develop a new executive order framework on cybersecurity and a "repeatable process for future model releases."

Availability and Pricing

Initial availability is limited to partners whose participation has been shared with the government. OpenAI plans to make GPT-5.6 models more broadly available to ChatGPT, Codex, and API users in the coming weeks.

OpenAI stated that Sol's pricing is nearly half that of Anthropic's Claude Fable 5 ($10 input / $50 output). Improved prompt caching is included to reduce costs for repeated prompts.