OpenAI has shown an unreleased model specialized for the cybersecurity field in public for the first time, sending ripples through the industry. During a livestream titled “Frontier at the Cyber” held in the early hours of November 13, 2026, the company abruptly revealed part of the performance of its next-generation cybersecurity-focused model. On “ExploitGym,” a benchmark that measures the ability to turn vulnerabilities into actually exploitable code, the next-generation model scored 77.0%. That is nearly 25 points above the 52.6% posted by the current specialized model, “GPT-6 Cyber.”
The broadcast first presented a comparison between the general-purpose frontier model “GPT-6 Astra” and GPT-6 Cyber, after which the next-generation model’s figure was added to the chart as a red line. The hosts explained that Anthropic’s “Mythos 5.2” (a higher-tier version of “Fable 5.2”), known for its strong overall capabilities, stays at roughly 38% on the same benchmark, emphasizing that the next-generation model reaches nearly double that level.
GPT-6 Cyber itself has already delivered results in the practical range. It discovered 81 previously unknown vulnerabilities lurking in the Firefox browser at a severity rating of “high,” and in a single run it examined roughly 1,000 vulnerabilities across Linux’s vast codebase, even generating fix patches. The next-generation model is positioned as an extension of that work.
| Model | ExploitGym Score |
|---|---|
| Next-generation cybersecurity-focused model (unreleased) | 77.0% |
| GPT-6 Cyber | 52.6% |
| Mythos 5.2 (Anthropic) | Approx. 38% |
Note: ExploitGym is a benchmark that measures the ability to implement vulnerabilities as exploitable code. The Mythos 5.2 figure is a reference value presented during the livestream.
OpenAI tends to keep high-performance models private. It was also an internal, unreleased model that generated a proof concerning the existence and smoothness of the Navier-Stokes equations, an unsolved problem in mathematics, but that model has not been made publicly available. Whereas GPT-6 Astra was launched in September, powerful models such as Google’s “Gemini 4 Carbon” and Anthropic’s “Claude Fable 5.2” appeared in quick succession over the following two months, and expectations for additional announcements toward year-end continue to smolder.
Behind the development race, meanwhile, the weakness of the revenue base is once again drawing attention. OpenAI told investors that its third-quarter operating loss reached $10.9 billion. The loss has widened from $10.1 billion in the previous quarter. Revenue grew about 50% from the second quarter but was not enough to offset enormous costs.
One of the drivers of revenue expansion is the advertising business. In February, the company began displaying ads in its chatbot. CEO Sam Altman had previously described advertising as an “unsettling” measure and positioned it as a “last resort,” but the company reversed course. As of August, it said the advertising division was expected to generate $1 billion in annual revenue.
Uncertainty over the revenue outlook also remains. In October, the company reported annualized revenue of about $50 billion, but that was $20 billion below the level it had shown investors a month earlier. The revision fueled doubts about the robustness of AI companies’ business models and contributed to concerns of a bubble.
Both OpenAI and Anthropic are eyeing historic initial public offerings (IPOs), and investors are watching with growing scrutiny. While Anthropic has cultivated corporate customers with coding tools, OpenAI had a strong consumer-oriented flavor, but its president said last week that the majority of sales now comes from corporate clients. Specialized models for the cybersecurity field are likely to be positioned as a trump card for capturing corporate demand.
Market participants are voicing the view that the next focal point will be whether the company can convert into revenue a model that overwhelms competitors on performance. For frontier companies that continue to make massive upfront investments, investors are paying closer attention than ever to the ability to achieve both growth and profitability.