OpenAI introduced a "limited preview" of three GPT-5.6 models on Friday: GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. The company also said it still plans to make the models "generally available in the coming weeks," but for now access is limited to "a small group of trusted partners in Codex and the API" at the request of the U.S. government.
In the accompanying model cards, OpenAI describes Sol as the "flagship model for ambitious agentic work," Terra as the "balanced model for efficient, everyday work," and Luna as the "fast, affordable model for high-volume work." The pricing listed alongside those cards puts Sol at $5.00 per input million tokens and $30.00 per output million tokens, Terra at $2.50 and $15.00, and Luna at $1.00 and $6.00, with lower cached-input rates for each.
OpenAI also highlighted benchmark claims for the line. A chart labeled TerminalBench 2.1 places GPT-5.6 Sol Ultra at 91.9%, Sol at 88.8%, Terra at 84.3% and Luna at 82.5%, with GPT-5.5 at 83.4%. The company further called Sol its "most capable model yet for cybersecurity," saying it improves long-horizon work tied to vulnerability research and exploitation.
Alongside those performance claims, OpenAI said Sol launches with its "most robust safety stack yet." The company mentioned real-time protections against "high-risk cyber activity and repeated misuse," plus human red teaming and more than 700,000 A100-equivalent GPU hours of automated testing. It also published an ExploitBench chart showing the family’s behavior across increasing output-token counts.
Source: OpenAI