GPT-6 Cyber: Critical-rated, $20/$100, Daybreak Red only

OpenAI launched GPT-6 Cyber on November 5, a security model its own system card rates Critical for cybersecurity, the top level of OpenAI's Preparedness Framework. It finds and exploits software vulnerabilities so defenders can test and fix them. It costs $20 per million input tokens and $100 per million output tokens. And you almost certainly can't get it: access runs through OpenAI's Daybreak Red program, with identity checks and, for individuals, a physical hardware key. The useful question for a small business isn't how to get GPT-6 Cyber. It's whether the people who guard your systems have it, and what they do with it.

What actually happened

From the GPT-6 Cyber system card, the API model page and VentureBeat:

• What it is. A variant of GPT-6 Astra with extra security training, built to respect authorization and stay inside the test scope, per VentureBeat.
• The rating. Critical in cybersecurity. VentureBeat reports GPT-5.6 Cyber was rated High.
• Who gets it. The system card says it's available "only through Daybreak Red to approved users and organizations." Organizations need SSO, MFA, managed devices and least-privilege access. Individuals need an invitation, identity and liveness checks, a legal attestation and at least one FIDO2 hardware key. API access is IP-allowlisted by default. VentureBeat says only selected Red customers get it at first.
• OpenAI's numbers. The system card lists 48.45% on ExploitGym against 42.43% for GPT-6 Astra. In a scope test, GPT-6 Cyber attacked out-of-scope services 0.07% of the time, against 5.51% for GPT-5.6 Cyber. These are OpenAI's own evaluations.
• The price. $20 input, $2 cached input and $100 output per million tokens, with a 1.05M-token context window. No Batch API. VentureBeat puts that at 60% more for input than GPT-5.6 Cyber, and double Anthropic's Claude Mythos 5.1 base rates.
• Hard limits. Extra blocking applies to industrial control systems, essential services, election systems and weapons systems.

Why GPT-6 Cyber matters for your business

Two things are true at once. The defenders with access just got a faster exploit finder. And OpenAI rating its own model Critical tells you how fast the attack side is moving: the same class of capability is what you'll face from people who don't fill out attestations. We've already seen what automated pipelines do at scale: OpenAI's earlier research produced 948 Linux kernel root exploits.

You won't run this model. Your MSP, your pentester or your platform vendor might. Ask them three things: do you use an AI exploit model, which one, and how fast do findings become patches? A vendor who finds bugs in hours but patches in quarters hasn't helped you.

On your side, the job is unchanged and more urgent: know what's exposed to the internet, patch it on a short clock, and remove what you don't use.

Sources: OpenAI GPT-6 Cyber system card, OpenAI API model page, VentureBeat.