OpenAI launches GPT-6 Cyber, a model for security research

OpenAI says it avoids exploiting out-of-scope services in 99.93% of cases, versus 94.49% for GPT-5.6 Cyber. On ExploitGym it scores 48.5%, against 32.1%. OpenAI rates its vulnerability-identification and exploit-development abilities as "critical", up from "high". Access is initially limited to approved Daybreak Red customers through the API and Codex. It is described as a specialized version of GPT-6 Astra. These figures come from the company's own evaluations.

Source: OpenAI, published on November 5, 2026.