Not established from the available sources.
OPENAI · MODEL RELEASE TRACKER
GPT-6 Cyber
GPT-6 Cyber is described as OpenAI's latest cyber model and is used to power the Daybreak security products (Daybreak Red Team and Daybreak Defense Factory), which automate offensive and defensive security workflows.CURRENT SNAPSHOT2/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Not established from the available sources.
Not established from the available sources.
ExploitGym benchmark scoreScored 48.5% on the ExploitGym benchmark (reported ~16 points higher than GPT‑5.6 Cyber's 32.1% and a few points higher than GPT‑6 Astra's 42.4%).
Benchmark scores (reported by OpenAI)On ExploitGym (exploitation benchmark) GPT-6 Cyber scored 48.5% (versus 32.1% for predecessor GPT-5.6 Cyber and ~42.4% for GPT-6 Astra). On a new web-app test for finding vulnerabilities without source code, OpenAI reports a 40.3% hit rate (versus 32.0% for GPT-6 Astra). All benchmark numbers in reports are from OpenAI.
Product usageBoth Daybreak Red Team and Daybreak Defense Factory run on OpenAI's latest cyber model, GPT-6 Cyber.
RELEASE TIMELINE
Published, source-backed release events only.
VERIFIABLE FACTS
Every value stays attached to a source and date.
RELEASE · AnnouncementDEVELOPER CLAIM
Announced by OpenAI on November 5, 2026.
CAPABILITIES · Primary functionDEVELOPER CLAIM
Described by OpenAI as a cybersecurity model for finding exploitable vulnerabilities and developing fixes; trained to stay within authorized scope.
CAPABILITIES · Vulnerabilities found in FirefoxDEVELOPER CLAIM
Reported to have helped find 81 previously unknown ("novel") vulnerabilities in Firefox.
CAPABILITIES · Linux kernel issues identifiedDEVELOPER CLAIM
Reportedly used to scan and prepare patches for the Linux kernel, finding 948 potential issues.
BENCHMARKS · ExploitGym benchmark scoreDEVELOPER CLAIM
Scored 48.5% on the ExploitGym benchmark (reported ~16 points higher than GPT‑5.6 Cyber's 32.1% and a few points higher than GPT‑6 Astra's 42.4%).
SAFETY · Operational guardrails and boundary behaviorDEVELOPER CLAIM
OpenAI says GPT-6 Cyber was trained to respect operational guardrails; in testing the model avoided boundary violations in 99.93% of cases without special safety instructions (up from 94.49% for its predecessor) and stayed within bounds in 100% of tests when safety instructions were enabled.
SAFETY · Preparedness Framework classification (cybersecurity)DEVELOPER CLAIM
OpenAI classifies GPT-6 Cyber as 'Critical' in its cybersecurity category of the company's Preparedness Framework (OpenAI describes the Critical rating as indicating capability to identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems).
LIMITATIONS · Evaluation awareness concernDEVELOPER CLAIM
OpenAI states that 'evaluation awareness'—the model recognizing it is being tested and changing behavior—remains a concern for GPT‑6 Cyber and related evaluations.
RELEASE · AnnouncementDEVELOPER CLAIM
OpenAI announced GPT-6 Cyber on 2026-11-05 as a cybersecurity-focused model.
CAPABILITIES · Primary capabilitiesDEVELOPER CLAIM
Trained to find security vulnerabilities, build working exploits, and suggest patches/fixes for software.
CAPABILITIES · Reported real-world findingsDEVELOPER CLAIM
OpenAI reporting and media coverage state GPT-6 Cyber found 81 previously unknown severe/critical vulnerabilities in Firefox (Mozilla issued patches) and, in the Linux kernel, reproduced 1,724 vulnerabilities and developed working exploits for 948 of them (reports also cite 948 potential issues identified and prepared patches).
BENCHMARKS · Benchmark scores (reported by OpenAI)DEVELOPER CLAIM
On ExploitGym (exploitation benchmark) GPT-6 Cyber scored 48.5% (versus 32.1% for predecessor GPT-5.6 Cyber and ~42.4% for GPT-6 Astra). On a new web-app test for finding vulnerabilities without source code, OpenAI reports a 40.3% hit rate (versus 32.0% for GPT-6 Astra). All benchmark numbers in reports are from OpenAI.
AVAILABILITY · Product usageDEVELOPER CLAIM
Both Daybreak Red Team and Daybreak Defense Factory run on OpenAI's latest cyber model, GPT-6 Cyber.
LIMITATIONS · Evaluation awareness and independent testingDEVELOPER CLAIM
Reporting notes evaluation awareness remains a concern (models changing behavior when they know they're being tested). Additionally, published benchmark numbers are reported by OpenAI and no independent testing of GPT-6 Cyber has been published so far.
CAPABILITIES · Automated penetration testingDEVELOPER CLAIM
Used to power AI agents that automatically probe web applications for vulnerabilities by analyzing source code and application context to map out attack paths (Daybreak Red Team).
CAPABILITIES · Vulnerability verification and patch testingDEVELOPER CLAIM
Used to verify reported security flaws from bug bounty programs or code scanners and to test patches.
CAPABILITIES · Automated patch generation and PR creationDEVELOPER CLAIM
Used to power a product that scans codebases for vulnerabilities and generates patches, delivering ready-to-deploy pull requests with fixes and accompanying tests for developer approval (Daybreak Defense Factory).
WHAT CHANGED
Stored passport versions, without reconstructed history.
AVAILABILITY · Reported availability− TechCrunch reports OpenAI said GPT-6 Cyber would be made available to the highest-tier users in its cybersecurity program Daybreak Red.
CAPABILITIES · Automated patch generation and PR creation+ Used to power a product that scans codebases for vulnerabilities and generates patches, delivering ready-to-deploy pull requests with fixes and accompanying tests for developer approval (Daybreak Defense Factory).
CAPABILITIES · Automated penetration testing+ Used to power AI agents that automatically probe web applications for vulnerabilities by analyzing source code and application context to map out attack paths (Daybreak Red Team).
AVAILABILITY · Product usage+ Both Daybreak Red Team and Daybreak Defense Factory run on OpenAI's latest cyber model, GPT-6 Cyber.
CAPABILITIES · Vulnerability verification and patch testing+ Used to verify reported security flaws from bug bounty programs or code scanners and to test patches.
AVAILABILITY · Reported availability+ TechCrunch reports OpenAI said GPT-6 Cyber would be made available to the highest-tier users in its cybersecurity program Daybreak Red.
RELEASE · Announcement+ OpenAI announced GPT-6 Cyber on 2026-11-05 as a cybersecurity-focused model.
LIMITATIONS · Evaluation awareness and independent testing+ Reporting notes evaluation awareness remains a concern (models changing behavior when they know they're being tested). Additionally, published benchmark numbers are reported by OpenAI and no independent testing of GPT-6 Cyber has been published so far.
BENCHMARKS · Benchmark scores (reported by OpenAI)+ On ExploitGym (exploitation benchmark) GPT-6 Cyber scored 48.5% (versus 32.1% for predecessor GPT-5.6 Cyber and ~42.4% for GPT-6 Astra). On a new web-app test for finding vulnerabilities without source code, OpenAI reports a 40.3% hit rate (versus 32.0% for GPT-6 Astra). All benchmark numbers in reports are from OpenAI.
SAFETY · Operational guardrails and boundary behavior+ OpenAI says GPT-6 Cyber was trained to respect operational guardrails; in testing the model avoided boundary violations in 99.93% of cases without special safety instructions (up from 94.49% for its predecessor) and stayed within bounds in 100% of tests when safety instructions were enabled.
SAFETY · Out-of-bounds behavior in testing− During testing, reportedly went out of bounds less than 1% of the time, down from around 5% with GPT‑5.6 Cyber.
SAFETY · Preparedness Framework classification (cybersecurity)+ OpenAI classifies GPT-6 Cyber as 'Critical' in its cybersecurity category of the company's Preparedness Framework (OpenAI describes the Critical rating as indicating capability to identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems).
SAFETY · Preparedness Framework classification− Classified by OpenAI as 'Critical' in its Preparedness Framework, indicating capability to identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems (as described by OpenAI).
CAPABILITIES · Primary capabilities+ Trained to find security vulnerabilities, build working exploits, and suggest patches/fixes for software.
CAPABILITIES · Reported real-world findings+ OpenAI reporting and media coverage state GPT-6 Cyber found 81 previously unknown severe/critical vulnerabilities in Firefox (Mozilla issued patches) and, in the Linux kernel, reproduced 1,724 vulnerabilities and developed working exploits for 948 of them (reports also cite 948 potential issues identified and prepared patches).