OpenAI has unveiled the GPT-5.6 model family, consisting of three distinct variants designed for different enterprise needs: Sol, Terra, and Luna. This marks a strategic shift from version-numbering to capability-tier naming.
Sol is the top-tier model for the most demanding tasks including complex reasoning, extended coding sessions, advanced agent-driven workflows, and security-focused applications. It introduces a new `max` reasoning setting for extended deliberation and an `ultra` mode with subagent-based parallel processing. Priced at $5.00/MTok input and $30.00/MTok output, Sol delivers the highest performance in the family.
Terra balances strong performance with efficiency for large-scale production environments like customer support and document analysis. At $2.50/$15 per million tokens, it's available for general API use.
Luna is the most cost-efficient option at $1.00/$6 per million tokens, optimized for speed and everyday tasks like summarization and routine automation. Despite being the cheapest, it performs near GPT-5.5 levels on several tests.
All three models crossed the "High" cyber threshold on internal capture-the-flag testing (Sol: 96.7%, Terra: 91.84%, Luna: 85.19%), classifying them at "High" risk for cyber and biological/chemical capabilities. Initial access is limited to approximately 20 vetted organizations following coordination with the US government under a new executive order framework. OpenAI publicly criticized this government gatekeeping process in its announcement.
Benchmark improvements include new state-of-the-art scores on TerminalBench 2.1 command-line tasks and ExploitGym, demonstrating measurable gains over GPT-5.5.