Google Launches Gemini 3.5 Flash Cyber to Compete in AI Security
Google has introduced Gemini 3.5 Flash Cyber, a specialized model designed to scan for software vulnerabilities via its CodeMender platform, positioning itself as a lower-cost alternative to Anthropic’s Mythos.
Google has launched Gemini 3.5 Flash Cyber, a specialized AI model engineered specifically for cybersecurity tasks. According to The Verge, the model is designed to integrate directly into CodeMender, Google's security-focused coding agent. By leveraging the underlying Gemini 3.5 Flash architecture, the system is intended to function as a cost-efficient alternative to larger, resource-intensive models currently dominating the sector, such as Anthropic’s Mythos 5. Google notes that CodeMender can invoke 3.5 Flash Cyber multiple times at high speed, enabling the agent to scan extensive code paths for vulnerabilities that might otherwise be missed.
The competitive landscape for AI-driven security is intensifying, with companies racing to match or exceed the capabilities of existing models. Anthropic’s Mythos 5, developed under the Project Glasswing initiative, is currently a benchmark for power but carries significant costs, reportedly twice as expensive as Claude Opus 4.8. Microsoft has already integrated Mythos into its security operations, recently completing its largest Patch Tuesday to date using AI-assisted vulnerability detection. Google’s entry into this space suggests a strategic pivot toward specialized, lower-cost models that can be deployed frequently without prohibitive compute expenses.
In terms of performance, Google claims that 3.5 Flash Cyber achieved competitive results against significantly larger models on the CyberGym AI cybersecurity benchmark when invoked up to five times. In a practical test involving the V8 JavaScript Engine, the model identified 55 unique confirmed issues. For comparison, Google reported that Gemini 3.5 Flash found 47 issues, while Opus 4.6 identified 36. Google further claimed that 3.5 Flash Cyber discovered 10 vulnerabilities that no other tested model could detect, suggesting that iterative invocation allows the model to uncover code paths that single-pass systems might overlook.
Access to the new model is currently restricted. Google stated that 3.5 Flash Cyber will be available first to governments and trusted partners through the CodeMender platform. This tiered rollout strategy reflects the sensitivity of security-focused AI tools, which require careful integration into existing infrastructure. While the company has not disclosed the full technical specifications of the model’s architecture or training data, the focus remains on operational utility—specifically the ability to perform high-frequency scanning at a lower price point than the current industry standard set by Mythos.
Alongside the security-focused release, Google also updated its broader Gemini lineup, including the introduction of Gemini 3.5 Flash-Lite, which the company describes as the most cost-effective model in the 3.5 series. Meanwhile, the standard Gemini 3.6 Flash has received upgrades to its coding and multimodal performance. These releases underscore a broader trend in the AI industry: moving away from monolithic, general-purpose models toward smaller, highly optimized systems tailored for specific enterprise and security workflows. Whether these specialized tools can maintain their performance edge as they are deployed at scale remains the primary question for security teams.
Sources
- 01 Google launches a cheaper alternative to large AI security models like Mythos — The Verge — AI