Google is launching a dedicated AI security model focused on the rapid identification and patching of software vulnerabilities. In a blog post published Tuesday, Google described Gemini 3.5 Flash Cyber as a "cost-efficient and highly capable alternative" to larger, more expensive AI systems, such as Anthropic’s Mythos.
Integration into CodeMender
The new model is built upon the Gemini 3.5 Flash architecture and will be initially available to governments and trusted partners via CodeMender, Google’s security-focused coding agent. Google notes that CodeMender can invoke 3.5 Flash Cyber multiple times at high speed and low cost, enabling AI agents to scan more code paths to uncover vulnerabilities.
Competition with Mythos
This launch positions Google against Anthropic’s Mythos 5, a powerful security model released under the Project Glasswing initiative. Mythos is compute-heavy and expensive, costing twice as much as Claude Opus 4.8. Microsoft, which adopted Mythos for its security protocols, recently recorded its largest Patch Tuesday after utilizing the AI to find vulnerabilities. Meanwhile, other competitors like China’s Z.ai claim their models can also compete with Mythos.
Performance Metrics
Google reports that 3.5 Flash Cyber achieved "competitive performance" against significantly larger models on the CyberGym AI cybersecurity benchmark. In testing against the V8 JavaScript Engine, the model identified 55 "unique confirmed issues," compared to 47 found by the standard Gemini 3.5 Flash and 36 by Opus 4.6. Notably, 3.5 Flash Cyber discovered 10 issues that no other model identified, continuing to find new vulnerabilities after being invoked multiple times.
