GPT-5.6 System Card [pdf] (deploymentsafety.openai.com)

OpenAI has previewed its new GPT-5.6 model family, which includes three variants: Sol, Terra, and Luna, each designed for different user needs ranging from flagship performance to cost-effectiveness and speed. This launch emphasizes robust safety features, marking a significant advancement in their cybersecurity capabilities while ensuring a commitment to broad access. The rollout will initially include a limited preview for selected partners due to government engagement, before making the models widely available. The significance of GPT-5.6 lies in its enhanced security measures and advanced training methodologies aimed at minimizing risks associated with its deployment. Notably, the models demonstrate improved capabilities in identifying vulnerabilities rather than exploiting them, providing an essential tool for defenders to fortify systems against potential threats. Enhanced safety protocols, including real-time monitoring during output generation, are tailored to each model's capabilities. The comprehensive testing and feedback mechanisms established for this release underline OpenAI's commitment to iteratively improving its AI systems while prioritizing user safety and responsible use.