Google launches Gemini 3.5 Flash Cyber for finding software bugs

In short: Google has launched Gemini 3.5 Flash Cyber, an AI tool meant to quickly find and help patch security problems in software.

What happened

Google says Gemini 3.5 Flash Cyber is a security focused version of its Gemini 3.5 Flash model. It is designed to spot security vulnerabilities, which are weak points in software that attackers can use, like an unlocked window.

The model will be offered first to governments and “trusted partners” through CodeMender, Google’s security focused coding agent. A coding agent is a tool that can read code and take actions, like scanning files and suggesting changes. Google says CodeMender can call the new model multiple times “at high speed and low cost,” which can help it check more possible routes through a program and uncover more issues.

Google is positioning this model as a cheaper option compared with larger and more expensive security models, including Anthropic’s Mythos. The Verge notes that Mythos 5 is considered powerful but costly to run.

Google also shared early benchmark results. It said Gemini 3.5 Flash Cyber performed competitively against much larger models on a cybersecurity test called CyberGym when it was allowed to try up to five times. Google also said the model found 55 unique confirmed issues in Google’s V8 JavaScript engine, compared with 47 found by regular Gemini 3.5 Flash and 36 by Opus 4.6, and that 10 of those issues were not found by other models.

Why it matters

More companies and governments are using AI to look for software weaknesses before criminals do. If tools like this get cheaper to use, more organizations may run these checks more often, which could lead to faster fixes and fewer security incidents.

Source: The Verge AI