Google DeepMind unveiled Gemini 3.5 Flash Cyber, a security-focused AI model designed specifically to find, validate, and patch software vulnerabilities. The model, built on top of the standard 3.5 Flash architecture, will not be available to the general public due to its dual-use nature as both a defensive and offensive tool.
The model runs within CodeMender, a multi-agent system that deploys parallel 3.5 Flash Cyber agents to produce combined vulnerability reports. Google says the cost-per-token advantage over larger models compounds into significantly more code coverage per session. The model reaches competitive performance on the CyberGym benchmark, the standard evaluation for AI-driven vulnerability discovery.
“AI models have become capable of finding security vulnerabilities faster than current systems can fix them,” Google stated in its announcement. The restricted deployment is designed to give defenders a head start through an exclusive pilot program for governments and trusted partners.
Gemini 3.5 Flash Cyber ships alongside two other model releases. Gemini 3.6 Flash delivers improved coding and knowledge work while consuming 17% fewer output tokens than 3.5 Flash, priced at $1.50 per million input tokens and $7.50 per million output tokens. Gemini 3.5 Flash-Lite runs at 350 output tokens per second at $0.30 per million input tokens, making it the fastest model in the 3.5 series for high-throughput production traffic.
