OpenAI has initiated a limited preview of its latest AI model series, GPT-5.6, introducing three distinct tiers: Sol, Terra, and Luna. The flagship model, Sol, is touted as the company’s most advanced and security-hardened AI to date. Access to Sol is currently restricted to a select group of trusted partners, following a formal request from the Trump administration, citing national security considerations.
The GPT-5.6 series adopts a new naming convention where the numerical identifier denotes the model’s generation, and the names Sol, Terra, and Luna represent different capability tiers. Sol stands as the premier model; Terra offers a balanced performance suitable for everyday tasks at half the cost of its predecessor, GPT-5.5; and Luna is designed for speed and affordability, providing robust AI capabilities at the lowest price point in the lineup.
Pricing for these models is structured per million tokens: Sol is priced at $5 for input and $30 for output, Terra at $2.50 for input and $15 for output, and Luna at $1 for input and $6 for output. Additionally, OpenAI plans to launch GPT-5.6 Sol on Cerebras in July, aiming to deliver inference speeds of up to 750 tokens per second for select enterprise clients.
The decision to limit the initial release of GPT-5.6 follows a formal request from the Trump administration, emphasizing the model’s advanced capabilities and potential national security implications. Reports indicate that the Office of the National Cyber Director and the Office of Science and Technology Policy, along with Commerce Secretary Howard Lutnick, advised against an unrestricted launch. OpenAI CEO Sam Altman informed employees that the government would be approving access on a case-by-case basis during the preview period. However, OpenAI has expressed that this government-controlled access is not intended to be a long-term approach, stating, “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.”
During this preview phase, GPT-5.6 models are accessible via the API and Codex to select trusted partners, with plans for broader availability across ChatGPT, Codex, and the API in the coming weeks.
GPT-5.6 Sol’s Cybersecurity Capabilities
GPT-5.6 Sol is engineered to excel in long-term security tasks, including vulnerability research and exploitation, marking a significant advancement in performance and efficiency. On the ExploitBench benchmark, Sol demonstrates competitiveness with Anthropic’s Mythos Preview while utilizing approximately one-third of the output tokens. Additionally, on ExploitGym—a benchmark developed collaboratively by UC Berkeley researchers, OpenAI, and other leading labs—Sol, Terra, and Luna all exhibit substantial improvements in cyber capabilities as reasoning efforts intensify.
In coding workflows, GPT-5.6 Sol sets a new standard on Terminal-Bench 2.1, achieving a score of 88.8%. The Sol Ultra configuration further elevates this benchmark to 91.9%, surpassing Claude Mythos 5 (84.3%) and GPT-5.5 (88.0%). The model also introduces a new ultra mode that leverages subagents to accelerate complex multi-step tasks beyond the capacity of any single agent.
Enhanced Cyberattack Protections
GPT-5.6 Sol is equipped with OpenAI’s most robust safety measures to date, specifically designed to withstand real-world adversarial attacks. This includes advanced safeguards against prompt injections and other manipulation techniques that have previously compromised AI models. By implementing these layered protections, OpenAI aims to ensure that GPT-5.6 Sol can be deployed securely in sensitive environments, addressing concerns about the potential misuse of powerful AI systems.
The introduction of GPT-5.6 Sol underscores OpenAI’s commitment to advancing AI capabilities while prioritizing security and responsible deployment. As AI models become increasingly integral to various sectors, the emphasis on robust safety measures and controlled access highlights the delicate balance between innovation and security in the AI landscape.
