OpenAI's new GPT-5.6-Cyber is built to fight AI-powered hacks
What's the story
OpenAI has announced the launch of GPT-5.6-Cyber, a specialized model designed for advanced vulnerability research and exploit development. The new addition is a fine-tuned version of OpenAI's most advanced general model, GPT-5.6 Sol, which was unveiled in June. The company claims that this new model can handle complex cybersecurity tasks like identifying zero-day vulnerabilities and creating exploit chains with an impressive success rate.
Benchmark results
GPT-5.6-Cyber achieved 95% on OpenAI's advanced cybersecurity benchmark
On the Advanced Cybersecurity Completion Rate benchmark, a test designed by OpenAI to assess tasks like exploit-chain development and privilege escalation, GPT-5.6-Cyber outperformed its immediate predecessor model, GPT-5.5-Cyber.
The new model achieved a completion rate of 95%, significantly higher than the 57.3% of its predecessor and way ahead of the general-purpose GPT-5.6 Sol model, which only managed a mere 1.5% with all safeguards applied.
Access restrictions
Access to GPT-5.6-Cyber limited to select enterprises
Despite its impressive capabilities, OpenAI has not made GPT-5.6-Cyber available to all ChatGPT or API customers.
Organizations wanting access must be accepted into a new tier of OpenAI's Daybreak cybersecurity program, called Daybreak Red. This tier provides access to dedicated cybersecurity models like GPT-5.6-Cyber.
The company has also introduced another new tier, Daybreak Blue, which gives a wider range of enterprises access to general models like GPT-5.6 Sol with some guardrails lifted for more cybersecurity uses.
Cost breakdown
GPT-5.6-Cyber more expensive than general models
OpenAI has priced GPT-5.6-Cyber at $12.5 per million input tokens and $75 per million output tokens, with cached input costing $1.25 per million tokens.
This makes it more expensive than GPT-5.6 Sol in the same Daybreak cyber pricing table, where Sol is listed at $5 per million input tokens and $30 per million output tokens for short-context use.
Program details
Daybreak Red for approved security teams only
Daybreak Red is meant for approved security teams doing advanced, authorized cyber work. This includes vulnerability research, penetration testing, red-team exercises, and exploit validation on systems the organization owns or has permission to test.
Enterprises wanting access have to apply through Daybreak Access, OpenAI's current pathway for vetting cyber users.
The application asks companies to identify who they are, what kind of security work they plan to do with these models, among other things.
Access criteria
OpenAI looking for serious security programs in applicants
OpenAI is looking for signs that the applicant has a serious security program of its own.
The company says participating enterprises need controls such as single sign-on, multifactor authentication, role-based access, employee-use monitoring, among others.
OpenAI also asks for a recognized security certification such as SOC 2 Type II or ISO 27001.
Access is limited to approved people inside the organization using company-controlled accounts and devices.