OpenAI launched GPT-6 Cyber on November 5, 2026. This specialized model for authorized vulnerability research identified 81 unknown high-severity flaws in the Firefox browser and created working exploits for 948 Linux kernel vulnerabilities.
GPT-6 Cyber Increases Exploit Success Rates
GPT-6 Cyber achieved a 48.5 per cent score on the ExploitGym benchmark. This result is 16 percentage points higher than the 32.1 per cent score recorded by GPT-5.6 Cyber, OpenAI stated. That figure represents a relative improvement of approximately 51% over the previous version.
The model also beat the general-purpose flagship GPT-6 Astra. That model posted a score of 42.4% on the same benchmark.
pic.twitter.com/PFHT1lhovt
In a black-box web penetration test, GPT-6 Cyber earned a score of 40.3%. GPT-6 Astra earned 32% on the same test evaluating vulnerability recall without application source code.
GPT-6 Cyber completed 95% of tasks in the RemoteExploitBench evaluation. GPT-6 Astra under Daybreak Blue refused all tasks in that same evaluation.
The model followed researcher-defined boundaries more reliably. During ExploitGym Honeypot testing, the model used services designed to be easy to exploit to see if it could recognize when it moved out of scope. GPT-6 Cyber avoided attacking out-of-scope systems in 99.93 per cent of tests without deployment safety prompts. It stayed within bounds 100 per cent of the time when those prompts were applied, SiliconANGLE reported.
GPT-5.6 Cyber went out of bounds in 5.51% of graded runs without deployment safety instructions.
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment. The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and… pic.twitter.com/34uAeNnhwP
With the 6 Cyber model, we were able to make huge improvements on this evaluation,
Fouad Matin, who leads OpenAI’s cyber preparedness and safety team, said during a press briefing, BigGo Finance reported.
Automated Linux Kernel Research Costs $1.2 Million
A research pipeline using the upcoming model targeted the Linux kernel in mid-October 2026. The system identified 2,308 candidate Linux kernel findings. It dynamically reproduced 1,724 vulnerabilities within 30 hours, The Hacker News reported. This scan targeted the most recent code in the kernel’s master branch.
🇺🇸 OpenAI unveils cyber model with unprecedented hacking skills – Axioshttps://t.co/R8i7kv7R8q
The pipeline produced 948 working local privilege escalation (LPE) exploits. These could allow an unprivileged local user to gain root access. This total is more than five times the 166 exploits produced by the earlier GPT-5.6 Sol model using the same research harness, BusinessLine noted. OpenAI reported that at least 10 of the techniques utilized by the model were previously unknown.
The automated run cost $1,213,430 in API credits, according to The Hacker News. This averaged $1,279 per LPE exploit. An earlier scan in late July used GPT-5.6 Sol and yielded 166 unique LPEs at a cost of $513,690.
Over 100 Linux LPEs were patched or discussed publicly within 10 days of the scan. Approximately 300 more are in progress.
OpenAI has disclosed 304 critical findings to maintainers along with candidate patches. The company is working with partners through its Patch the Planet initiative to send remaining fixes in stages. OpenAI warned that for the near and medium term, having access to a low-privileged account on an unpatched operating system should be considered functionally equivalent to root access for a well-resourced threat actor.
Our security researchers built an autonomous cyber vulnerability discovery, exploitation, and patching pipeline. In a single run with an upcoming cyber model, it validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel at a total API… pic.twitter.com/Nu2I0n2ayI
Mozilla Fixes Firefox Vulnerabilities
The model uncovered 81 previously unknown, high-severity vulnerabilities in the Firefox web browser. OpenAI shared evidence, reproducible tests, and proposed patches with Mozilla. Mozilla has already begun incorporating fixes for many of these flaws, BusinessLine reported. The model also identified 30 known Firefox issues that were more severe than previously reported.
GPT-6 Cyber is a world-class model for cyberdefense, Brian Grinstead, a Firefox distinguished engineer at Mozilla, said in a statement, BigGo Finance reported. Grinstead added that the model surfaced previously unknown high-severity flaws and architectural weaknesses even after Firefox had been hardened through multiple generations of frontier models, helping the browser maker push fixes to millions of users.
OpenAI Restricts Early Access
OpenAI is limiting initial access to GPT-6 Cyber to approved customers within its Daybreak Red program. This program is designed for offensive-style tasks like penetration testing and red teaming, Marketscale reported. OpenAI is also expanding its Daybreak Blue alpha access for GPT-6 Astra to more organizations and individual security researchers.
The company plans to expand access to all approved Daybreak Red customers in the coming weeks. OpenAI is making the older GPT-5.6 Cyber available to all customers in the Daybreak Red tier, Thedigitalbriefnews reported. OpenAI has also created a separate subgroup for approved critical infrastructure defenders to conduct authorized security testing.
GPT-6 Cyber costs 60% more for input and approximately 33% more for output than GPT-5.6 Cyber.
| Model | Input (per 1M) | Output (per 1M) | Cached Input (per 1M) |
|---|---|---|---|
| OpenAI GPT-6 Cyber | $20 | $100 | $2 |
| OpenAI GPT-5.6 Cyber | $12.50 | $75 | $1.25 |
| Anthropic Claude Mythos 5.1 | $10 | $50 | $0.25 |
Earlier coverage: OpenAI's GPT-6 Cyber finds 948 Linux kernel vulnerabilities