GPT-5.6 Cyber

OpenAI released GPT-5.6 Cyber on August 10, 2026, built on top of GPT-5.6 Sol and trained specifically to help defenders find and fix vulnerabilities faster than attackers can exploit them. It sits behind Daybreak Red, a new, more restricted tier of OpenAI’s Daybreak cyber defense program that requires separate vetting beyond the general Daybreak Blue access most approved security teams get. Compared with the base GPT-5.6 Sol model, GPT-5.6 Cyber has far fewer refusals on dual-use security requests, completing 95 percent of advanced tasks on OpenAI’s internal cybersecurity evaluation versus 1.5 percent for the general model.

OpenAI says the model was used internally to find two previously unknown vulnerabilities in V8, Chrome’s JavaScript engine, which Google has since patched. Pricing runs well above the standard GPT-5.6 line, at $12.50 per million input tokens and $75 per million output tokens, reflecting both the specialized training and the access controls OpenAI has wrapped around it. The company frames the release as a response to AI-assisted attacks becoming more common, arguing that defenders need equivalent capability rather than relying on general-purpose models with safety guardrails that also slow down legitimate vulnerability research.