OpenAI Unveils GPT-5.6 Suite and First Offense‑Grade Cyber Model in Limited Preview

OpenAI announces release of GPT-5.6 series, naming Sol, Terra, Luna, each tuned for scientific reasoning, multilingual tasks, and creative generation respectively, available only to a select group of preview partners under US government oversight. Models build on GPT-5 architecture with enhancements for specialized tasks.

GPT‑5.6‑Cyber is described as OpenAI’s first “offense‑grade” hacking model, trained on red‑team datasets to mimic adversarial techniques while incorporating safety layers, designed to simulate cyber‑attack scenarios for defensive research and to help organizations test vulnerabilities.

Regulators in the United States are reviewing the launch amid growing concerns over AI‑driven threats and potential misuse, prompting OpenAI to limit access, implement monitoring tools, and cooperate closely with oversight bodies to ensure responsible deployment.

Agent Plugins standard is updated alongside the GPT‑5.6 rollout, adding real‑time data retrieval and tool‑use capabilities, enabling developers to integrate custom functions, reflecting OpenAI’s push to expand functional extensions as the model marks its first anniversary.