OpenAI updates GPT 5.5 Cyber AI model, claims it outperforms Mythos 5 on important benchmark
Follow UsOpenAI has announced an updated version of GPT 5.5 Cyber.
GPT 5.5 Cyber is said to be OpenAI's strongest model yet for finding and helping patch software vulnerabilities.
According to OpenAI, GPT 5.5 Cyber outperforms Mythos 5 on CyberGym benchmark.
OpenAI has announced it is releasing the full version of GPT 5.5 Cyber, its AI model designed for cybersecurity work. According to the company, the new version is “more permissive” and “more capable for advanced, authorised cybersecurity work.” The updated GPT 5.5 Cyber can do much more than identify software vulnerabilities. It can analyse large codebases, find security related components, check if vulnerable code is reachable and test possible fixes.
SurveyGPT 5.5 Cyber is said to be OpenAI’s strongest model yet for finding and helping patch software vulnerabilities. It also retains GPT 5.5’s general-purpose intelligence and ability to work across long, complex tasks, according to the company.
Also read: Nothing Phone 4b in the works? Company hints at new lineup, here is what we know
The company shared benchmark results showing improvements over the standard GPT 5.5 model. On CyberGym, a benchmark that measures whether an AI agent can reproduce known software vulnerabilities, GPT 5.5 Cyber scored 85.6 per cent. GPT 5.5 scored 81.8 per cent on the same test. The new model even outperformed Anthropic’s Mythos 5 on the CyberGym benchmark, according to the comparison shared by OpenAI.
On ExploitGym, which measures whether an AI can turn known vulnerabilities into working exploits that achieve unauthorised code execution, GPT 5.5 Cyber achieved 39.5 per cent. For comparison, GPT 5.5 scored 25.95 per cent.
Also read: Got GTA 6 beta invitation? It can be a scam designed to steal your information
OpenAI said benchmark scores are only one part of the story. “What matters in practice is whether a model can find real vulnerabilities, distinguish actionable issues from noise, and help defenders land fixes safely. We are continuing to evaluate the model’s performance on complex repositories and real remediation workflows as coordinated disclosures conclude,” the company added.
The company also said it has been working closely with the US government on cybersecurity and AI safety. This includes collaboration with the Center for AI Standards and Innovation (CAISI), the Office of the National Cyber Director (ONCD), and the Office of Science and Technology Policy (OSTP).
According to OpenAI, GPT 5.5 Cyber is intended for verified defenders who need advanced cybersecurity capabilities. For most users, the company recommends GPT 5.5 with Trusted Access.
Also read: Anthropic and US govt in talks to create AI security framework after Claude Fable 5 ban: Report
