OpenAI has introduced its latest GPT-5.6 series of AI models, yet the most advanced versions will remain unavailable to nearly all users for now. The lineup features three variants: Sol as the top-performing model, Terra as a balanced mid-level option, and Luna as a quicker, lower-cost choice. Sol leads in benchmarks covering cybersecurity, biology, and autonomous tasks, and incorporates multiple safety layers to reduce risks of misuse in cyberattacks or other harmful activities. Release to the public has been postponed indefinitely at the request of the US government. Only a limited group of pre-approved customers may access the models initially. OpenAI is coordinating with the administration to broaden availability starting next week, though approval standards remain unclear. This marks the second instance of a leading AI company restricting its newest models due to government directives. The move has prompted debate over regulatory influence on AI development. OpenAI expressed hope for wider access soon while noting the current process is not ideal for long-term use. Sol matches prior pricing, while Terra offers comparable performance at reduced cost. The models are named after celestial terms, with Sol focused on enhanced agent capabilities in coding and security tasks. Safety measures are integrated into the core design to prioritize defensive applications.
