The VentureBeat report describes GPT-6 Cyber as OpenAI’s latest specialized model for authorized vulnerability research, exploit development and security investigations. The product is entering a market where Anthropic and Google also offer cybersecurity-focused systems, but OpenAI is positioning its model at the more aggressive end of the spectrum.
That positioning matters commercially. Security buyers do not necessarily need another model that can explain a vulnerability or suggest a patch. They may pay more for a system that can reproduce a crash, determine whether it is exploitable and produce a working proof of concept. The question is whether those capabilities reduce enough analyst time and operational risk to justify OpenAI’s premium.
A model designed to do more than identify flaws
OpenAI said in its launch announcement that GPT-6 Cyber is built for authorized vulnerability research, exploit development and web application penetration testing. The company says the model can reproduce crashes, construct exploits and operate within the limits of a defined assignment.
OpenAI reports a 48.5% score on its ExploitGym evaluation, compared with 32.1% for GPT-5.6 Cyber. The announcement also cites strong results in web application testing and a separate RemoteExploitBench evaluation. Those measurements are intended to show that the model is not simply better at describing security problems. It is better at carrying out the multi-step work required to validate them.
The company says researchers used GPT-6 Cyber to reproduce 1,724 Linux kernel vulnerabilities and develop working exploits for 948 of them. It also reports 81 high-severity Firefox findings. These figures are central to OpenAI’s commercial argument because they connect benchmark performance with work that security research teams already perform manually.
Mozilla provides some independent context. Mozilla’s Firefox security advisory records numerous high-severity vulnerabilities reported by OpenAI Codex Security. The advisory supports the existence of OpenAI-assisted Firefox findings, but it does not confirm OpenAI’s broader claim of 81 vulnerabilities. That distinction is important when evaluating company-reported performance.
Access is restricted because the capability is powerful
OpenAI is not releasing the model as an unrestricted general-purpose API. OpenAI’s system card classifies GPT-6 Cyber as a Critical cybersecurity system and places it inside the gated Daybreak Red program.
The system card describes restrictions, evaluation methods, safety mitigations and alignment results. In practical terms, OpenAI is trying to sell the model’s offensive capability while limiting who can use it and what assignments it can perform. The defined-scope approach is designed to reduce misuse, but it also creates friction for customers that want immediate, broad access.
That tradeoff could shape adoption as much as raw model performance. Large security vendors may accept a review process if the model materially increases the output of expert teams. Smaller consultancies, independent researchers and companies with urgent testing needs may find the restrictions less attractive, particularly if competing products are easier to access.
The limits also protect OpenAI from one of the most serious risks in the cybersecurity market. A model capable of creating working exploits can help defenders validate patches, but the same capability could lower the barrier to attacking unpatched systems. Daybreak Red therefore functions as both a safety mechanism and a market filter.
The premium has to produce measurable returns
OpenAI’s API documentation lists GPT-6 Cyber at $20 per million input tokens and $100 per million output tokens. The output price is especially significant because exploit development and investigation can require long, iterative responses that include code, test results and troubleshooting.
Those rates place GPT-6 Cyber substantially above the cyber offerings from Anthropic and Google, according to the VentureBeat comparison. That pricing gives OpenAI a clear strategic position: it is not competing primarily on affordability, but on the value of higher-end security work.
For a security operations center, the relevant calculation is not the cost per token alone. It is whether the model can reduce the time required to triage a vulnerability, reproduce a failure or produce evidence that a flaw is exploitable. If a costly model replaces hours of specialist work, the economics can still be favorable. If analysts must repeatedly review incomplete exploits or correct unsafe actions, the premium becomes harder to defend.
OpenAI’s reported Linux and Firefox results offer a basis for that argument, but they remain company claims with limited public comparison data. The benchmarks may indicate a meaningful capability lead, yet buyers will need evidence from production environments, including false-positive rates, containment requirements and the amount of human supervision needed.
OpenAI’s advantage is capability, not reach
GPT-6 Cyber gives OpenAI a potentially valuable foothold in high-end cybersecurity. Its advantage is the ability to market a model that works closer to the point where vulnerability research becomes operational. Its weakness is that the same capability demands restrictions, higher oversight and higher prices.
No opposing commercial view was provided in the supplied sources, and there is no reported response from Anthropic or Google challenging OpenAI’s performance claims or pricing strategy. For now, the market test will come from authorized users in the Daybreak Red program.
If those users show that GPT-6 Cyber can consistently turn difficult security investigations into faster, defensible results, OpenAI may establish a premium category for offensive security AI. If the practical gains are narrower than the benchmarks suggest, customers may prefer cheaper and more accessible alternatives. The model’s competitive advantage will ultimately depend less on its highest exploit count than on whether security teams can convert that capability into reliable savings and better protection.
Rebeca Smith is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Rebeca Smith's usual length and in Rebeca Smith's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.