GPT-6 Cyber is a cybersecurity model OpenAI released on November 5, 2026, trained to find and validate exploitable vulnerabilities while staying within the scope researchers set. It succeeds GPT-5.6 Cyber, scores 48.5% on ExploitGym (GPT-5.6 Cyber: 32.1%), and is the first OpenAI model rated Critical in cybersecurity under its Preparedness Framework.

Who can use GPT-6 Cyber?

Initially only a select group of approved Daybreak Red customers, through the API and Codex. Daybreak Red is the tier of OpenAI's Daybreak security program for authorized offensive work such as exploit development and penetration testing. OpenAI says it will expand access to all Daybreak Red customers soon and is making GPT-5.6 Cyber available to all Daybreak Red customers now. New organizations apply through openai.com/daybreak. API access requires IP allowlisting by default.

How good is GPT-6 Cyber compared with GPT-6 Astra?

OpenAI says GPT-6 Cyber outperforms GPT-6 Astra on most cyber tasks even when Astra does not refuse. On its new web penetration-testing evaluation it reached 40.3% recall of vulnerabilities against 32.0% for Astra, and on RemoteExploitBench it completed 95% of tasks while Astra running under Daybreak Blue refused all of them. In OpenAI's research it developed working exploits for 948 Linux kernel vulnerabilities and found 81 new high-severity Firefox flaws.

What is GPT-6 Cyber's price?

OpenAI has not published a public per-token price for GPT-6 Cyber; access is through approved Daybreak Red agreements. For general code review and vulnerability triage, OpenAI points to mainline models such as GPT-6 Astra, and to Daybreak Blue for verified defenders who need fewer refusals.

Quick Answer

What Is OpenAI GPT-6 Cyber? Access, Benchmarks, Limits

Published: November 8, 2026

The short answer

GPT-6 Cyber is OpenAI’s specialist model for offensive security research, released November 5, 2026. It is trained to find vulnerabilities, write and validate working exploits, and stay inside the rules of engagement a researcher sets. It is not generally available: access starts with a select group of approved Daybreak Red customers through the API and Codex.

Key facts

	GPT-6 Cyber
Released	November 5, 2026
Predecessor	GPT-5.6 Cyber
Preparedness rating	Critical in cybersecurity (GPT-5.6 Cyber: High)
ExploitGym	48.5% (GPT-5.6 Cyber 32.1%)
Web pentest recall (new OpenAI eval)	40.3% (GPT-6 Astra 32.0%)
RemoteExploitBench	95% completed (GPT-6 Astra with Daybreak Blue refused 100%)
Stays in scope (ExploitGym Honeypot)	99.93% without deployment prompt, 100% with it (GPT-5.6 Cyber 94.49%)
Access	Select Daybreak Red customers, API + Codex; IP allowlisting required by default
Public price	Not published

All figures are OpenAI’s, from its November 5, 2026 announcement and system card.

What it has done

OpenAI reports that GPT-6 Cyber reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 — more than five times the 166 GPT-5.6 Sol managed with the same harness — and that 304 critical findings went to maintainers with candidate patches. It produced evidence and reproducible tests for 81 previously unknown high-severity Firefox vulnerabilities, and Mozilla has shipped fixes for many of them. Security teams at Cisco, Calif and CodeRabbit use it for vulnerability research and code review.

Why “staying in scope” is the headline

A security agent that wanders outside its target is a liability. OpenAI built ExploitGym Honeypot, which plants easy-to-exploit services next to the real target; exploiting them is out of scope. GPT-6 Cyber left them alone in 99.93% of runs without its deployment prompt and 100% with it. The test was informed by an earlier incident involving Hugging Face.

Safeguards around it

Cyber-specific misalignment monitoring checks that actions match the user’s authorization.
Targeted monitoring for emerging threats, such as exploit attempts against live PHP servers during a coordinated disclosure.
Private Safety Processing for Zero Data Retention API customers, reviewing encrypted records kept in customer storage for 30 days.
IP allowlisting for Daybreak Red API access by default, plus credential-compromise monitoring.
New checks against high-risk activity on critical infrastructure (OT/ICS), refined with Dragos.

How to get access

Defenders doing code review and triage: start with GPT-6 Astra; OpenAI says mainline models already support review, triage and patch development.
Verified defenders who hit refusals: apply for Daybreak Blue, now expanding to more organizations and individual researchers.
Authorized offensive work (exploit development, pentesting, red teaming): apply for Daybreak Red at openai.com/daybreak. Existing Daybreak Red customers get GPT-5.6 Cyber now and GPT-6 Cyber as access expands.

For how this compares with other labs’ security models, see best AI models for cybersecurity defense and our explainer on OpenAI Daybreak.

Last verified: November 8, 2026.