OpenAI Previews GPT-5.6 Sol With Restricted Access

OpenAI released three versions of its GPT-5.6 model on Friday: Sol, Terra, and Luna. These models are currently available as a limited preview to a select group of companies, including participants in an ongoing engagement with the U.S. government. The Sol model represents OpenAI's latest flagship and most powerful offering. The Terra model is designed to provide a balance between computational efficiency and performance, while the Luna model has been specifically fine-tuned for enhanced speed and affordability.

GPT-5.6 Sol is positioned as the most advanced iteration, incorporating enhanced capabilities and potentially new features not detailed in the announcement. The strategic release to a limited set of partners suggests a phased rollout strategy, allowing OpenAI to gather feedback and refine the models before a broader public release. This approach is common for major AI model launches, enabling the company to address potential issues and optimize performance based on real-world usage scenarios.

The inclusion of U.S. government entities in this preview phase indicates a focus on applications relevant to national security, research, and public sector initiatives. OpenAI has previously collaborated with government agencies on various AI projects, and this preview likely serves to explore further integration and utility of their advanced language models within governmental frameworks. The specific nature of these collaborations and the intended use cases remain undisclosed.

While the exact technical specifications and benchmark performance data for Sol, Terra, and Luna were not fully detailed in the announcement, the tiered release strategy implies distinct performance profiles for each model. Sol is expected to lead in complex reasoning and generation tasks, Terra in cost-effective deployment for moderate workloads, and Luna in applications requiring rapid response times and lower operational costs. OpenAI has not yet provided a timeline for general availability of these GPT-5.6 models.