OpenAI previews GPT-5.6 Sol, Terra, and Luna in limited release

OpenAI announced a limited preview of its GPT-5.6 model family on June 26, 2026, introducing three models with distinct capability and cost profiles. Sol is the flagship model, positioned as the most capable and specifically optimized for cybersecurity and long-horizon coding tasks. Terra is designed for standard business workflows where Sol’s ceiling is not required. Luna is the fast, high-volume tier for latency-sensitive or cost-sensitive applications.

The preview launched with access limited to approximately 20 partner organizations. OpenAI shared its release plans with the US government before the launch, and reporting from June 25 indicated the White House had asked the company to proceed carefully with the rollout. General availability for all three models was expected in the weeks following the preview.

Sol includes what OpenAI describes as its most robust safety stack to date, with reinforced protections for higher-risk requests, particularly in cybersecurity and repeated misuse scenarios. The model sets a new state-of-the-art on Terminal-Bench 2.1, a benchmark for long-horizon terminal tasks.

For product teams, the three-tier model architecture has immediate planning implications. Each model in a family requires a separate integration decision: which tasks benefit from Sol’s capability ceiling, which run adequately on Terra, and which are better served by Luna’s speed and cost profile. As frontier labs release model families rather than single models, this tier-selection decision becomes part of product and pricing strategy—not just an engineering choice. Teams building AI features need roadmap clarity on which tier they are building against and how they will handle future tier migrations.

GPT-4.5 was simultaneously deprecated from ChatGPT as part of the same announcement.