OpenAI GPT-5.6 Sol, Terra, Luna: Limited Preview & Security

OpenAI has launched a limited preview of its new GPT-5.6 model family, which includes three variants: Sol, Terra, and Luna. This update aims to enhance capabilities in areas like software engineering, scientific research, and cybersecurity. Here's the thing: Access to these models is restricted to a small group of trusted partners. This is due to ongoing coordination with the U.S. government, which has requested limitations on the initial release over security concerns. This marks the first time the U.S. government has made such a request to an American AI company for a frontier model. The GPT-5.6 family offers different tiers for intelligence, speed, and cost. Sol is the most capable, designed for complex tasks. Terra balances performance with cost, while Luna is the fastest and most cost-efficient. What's interesting is Sol features improved agentic capabilities, especially in coding, biology, and cybersecurity. It also introduces new reasoning controls, like a "max" reasoning effort mode for deep problem-solving and an "ultra" mode for complex, multi-step workflows. Benchmark results show Sol's advancements. On Terminal-Bench 2.1, Sol Ultra achieved a state-of-the-art score of 91.9%. The model also performs well on ExploitBench and shows gains on ExploitGym for cybersecurity tasks. In biology, Sol delivers stronger results on GeneBench v1. The bottom line is these new models offer significant advancements, but their controlled release highlights the growing importance of security in AI development.