Google unveils cost-efficient AI security model to rival Anthropic’s Mythos
The new model, built on Gemini 3.5 Flash and integrated with CodeMender, is initially available to governments and trusted partners, offering a cheaper alternative to heavy AI security systems.
Google has launched Gemini 3.5 Flash Cyber, a new artificial intelligence model specifically designed for cybersecurity tasks. Built on the Gemini 3.5 Flash architecture, the system aims to identify and patch software vulnerabilities at a lower cost than larger, compute-heavy models such as Anthropic’s Mythos. The model is initially available to governments and trusted partners through CodeMender, Google’s security-focused coding agent.
Google describes Gemini 3.5 Flash Cyber as a cost-efficient alternative to expensive AI security systems. The model integrates with CodeMender, which can invoke the AI multiple times at high speed and low cost. This capability allows security agents to scan more code paths and uncover vulnerabilities that might otherwise be missed in a single pass.
In testing against the CyberGym AI cybersecurity benchmark, Google states the model achieved competitive performance compared to significantly larger systems when invoked up to five times. The model identified 55 unique confirmed issues in the V8 JavaScript Engine. This outperformed the standard Gemini 3.5 Flash, which found 47 issues, and Anthropic’s Opus 4.6, which identified 36.
Google noted that Gemini 3.5 Flash Cyber discovered 10 issues that no other model found. The company highlighted that the model continued to identify new code paths and vulnerabilities after multiple invocations, suggesting a persistent scanning capability that adds depth to automated security checks.
The launch occurs amid a broader industry race to develop affordable AI security tools. Microsoft recently adopted Anthropic’s Mythos for its security checks, resulting in its largest Patch Tuesday this month. Meanwhile, China’s Z.ai has claimed its own model can compete with Mythos. Anthropic’s Mythos 5, released as part of Project Glasswing, is described as compute-heavy and expensive, costing twice as much as Claude Opus 4.8.
Alongside the security model, Google introduced Gemini 3.6 Flash, which offers improvements in coding and multimodal performance. The company also released a new 3.5 Flash-Lite model, positioning it as the most cost-effective option in the 3.5 series.
Google’s blog post on Tuesday outlined these releases, emphasising the strategic shift towards accessible AI security. The move positions Google to compete more directly in the enterprise and government sectors, where cost and efficiency are critical factors in adopting advanced AI tools.
