Google DeepMind Unveils Gemini 3.5 Flash Cyber — a Lightweight Model for Cybersecurity

Google DeepMind has introduced Gemini 3.5 Flash Cyber, a lightweight cybersecurity model built on 3.5 Flash. The model is tailored for rapid and efficient discovery, confirmation, and remediation of vulnerabilities, outperforming the main Flash line in these tasks. Leveraging its speed and low cost, it is suitable for frequent scanning of codebases, including during release processes and commit reviews. 3.5 Flash Cyber will be available through a limited pilot program for governments and trusted partners via the CodeMender tool. In benchmarks such as CyberGym and Big Sleep tests, the model achieves results comparable to or surpassing larger models. In a production scan of Chrome, it identified more unique vulnerabilities than the base 3.5 Flash and Claude Opus 4.6, including issues missed by other models. At Google, the model is already used to protect internal codebases for Chrome, Android, Cloud, Ads, and YouTube, finding vulnerabilities within hours and generating exploits that bypass standard defenses like ASLR (Address Space Layout Randomization) and W^X (Write XOR Execute). Positive feedback has been received from testers at Wiz and Cloud CISO Security Engineering.