OpenAI previews GPT-5.6 Sol with restricted access and cyber safeguards

OpenAI released three GPT-5.6 models—Sol, Terra, Luna—in a limited preview for select companies under a U.S. government engagement. Sol, the flagship, features the most robust safety stack and matches Anthropic Mythos Preview on ExploitBench with one-third the tokens. The preview enforces guardrails to block offensive cyber activity while permitting defensive security work.