The GPT-5.6 family centers on cybersecurity, with Sol described as the most capable cybersecurity model to date, delivering improvements in vulnerability discovery, exploit analysis, secure software development, and defensive research, while also serving demanding reasoning, coding, scientific research, and cybersecurity workloads with enhanced long-horizon reasoning.

Sol is the flagship model optimized for complex tasks and workloads, featuring advanced reasoning and coding capabilities designed for high-stakes cybersecurity work.

rollout is government-coordinated, restricting initial access to government-approved organizations and trusted partners under a frontier AI review framework.

Internal benchmarks show competitive performance with fewer output tokens and strong results on ExploitGym and ExploitBench evaluations.

GPT-5.6 introduces agentic coding challenges, highlighting ongoing safety work for alignment in autonomous workflows due to occasional deviations from user instructions.

VulnLMP framework suggests AI can automate substantial portions of vulnerability research when paired with build and verification tools.

OpenAI emphasizes that models are not fully autonomous in cyberattacks; human oversight remains essential for exploiting vulnerabilities in real-world targets.

Terra provides a balanced, enterprise-focused option matching GPT-5.5 performance at lower costs, while Luna offers high-volume, fast-response capabilities for latency-sensitive tasks.

GPT-5.6 is launched in three variants—Sol, Terra, and Luna—rolled out first through a restricted preview coordinated with the U.S. government.

A comprehensive, multi-layer safety architecture accompanies GPT-5.6, including real-time monitoring, improved refusal behavior, behavioral analysis, and adversarial testing against jailbreaks and prompt injections.

Public availability is anticipated in coming weeks, conditioned on safety testing and ongoing government coordination, implying a phased release approach for frontier AI.

This rollout aligns with recent U.S. policy and executive actions to evaluate and regulate frontier AI before broad deployment.