OpenAI Launches Limited Preview of GPT-5.6 Family

OpenAI has initiated a limited preview of the new artificial intelligence model "GPT-5.6" family. This family comprises three models with different applications—Sol, Terra, and Luna—and is currently available to only approximately 20 organizations. OpenAI stated that general availability is planned "within weeks."

The three models are each designed for different purposes. Sol is the top-tier model intended for tasks requiring advanced reasoning, such as complex coding and security research. Terra addresses business operations requiring high-volume processing, including customer support, internal tools, and document analysis. Luna, meanwhile, is designed to quickly and cost-effectively handle everyday tasks such as document summarization, draft creation, and automation of routine operations. While Sol and Terra achieved high benchmark scores, Luna, despite its fastest and lowest-price positioning, demonstrated performance comparable to GPT-5.5 across multiple tests.

Regarding pricing, Sol's rates have been confirmed to match GPT-5.5 levels at 5 dollars per one million input tokens and 30 dollars per one million output tokens. However, pricing details for Terra and Luna are not currently specified in the source material, and exact figures remain undisclosed. OpenAI stated that Sol realizes significant performance improvements in long-duration coding and security-related tasks.

This phased release is deeply influenced by U.S. policy developments. President Trump issued an executive order on June 2, 2026, instructing the establishment of an interagency federal framework to evaluate the safety and social appropriateness of new artificial intelligence models. This review process is scheduled to proceed over 30 days, with July 2 as the target date. During this period, OpenAI explained that it pre-shared plans and model performance with the U.S. government and chose to provide access only to a limited number of trusted partner companies at the government's request.

This movement is also influenced by developments regarding competitors. Anthropic, a major competitor of OpenAI, received an export control order from the U.S. government after jailbreak techniques were discovered in its powerful model "Claude Fable 5." In response, Anthropic suspended all access to Fable 5 and its cybersecurity-specialized model "Claude Mythos 5" for both general and enterprise users. Additionally, Anthropic had previously provided limited access to the predecessor model of Mythos 5 as "Claude Mythos Preview" to participants in the cybersecurity research program "Project Glasswing," dating back to April of the same year.

This series of developments can be interpreted as marking a transition to a new era in which the public release process for artificial intelligence models is linked with government safety reviews. Historically, artificial intelligence companies have often released new models based on independent judgment, but going forward, a trend toward phased releases coordinated with government reviews and regulatory frameworks may become established. For corporate adoption managers, it is important to note that real-time safety response and regulatory compliance are becoming increasingly critical prerequisites for artificial intelligence adoption.

The general availability date for the GPT-5.6 family remains unspecified at this time, with only "within weeks" indicated. As the government-led evaluation process proceeds with July 2 as a milestone, how its results affect the release schedule will be a key focus going forward. As artificial intelligence industry-wide release practices evolve, future developments may influence the actions of other companies as well.