source: google deepmind: introducing gemini 3.5 flash cyber
level: technical
google deepmind released gemini 3.5 flash cyber, a cybersecurity model built on the 3.5 flash architecture and fine-tuned to find, validate, and patch software vulnerabilities. the model is designed to be cost-efficient and scalable, addressing the need for defenders to keep pace with ai-driven vulnerability discovery. it will be available through a limited-access pilot to governments and trusted partners via codemender, google's code security agent.
the model's lightweight design allows it to scan large codebases and analyze many code paths quickly. codemender can invoke 3.5 flash cyber multiple times to produce a single high-quality report, making it suitable for frequent scans and commit pipelines. on the cybergym benchmark, the agent achieved competitive performance against larger models by using up to five calls per report. in internal tests on chrome and safari, it outperformed mainline 3.5 and 3.6 flash models.
in real-world use at google, 3.5 flash cyber found remote code execution and memory-corruption vulnerabilities in production services within hours. it also discovered 55 unique issues in the v8 javascript engine, more than mainline 3.5 flash and claude opus 4.6. early testers from wiz and cloud ciso security engineering confirmed significant capability improvements over the base 3.5 flash model.
why it matters: this model gives security teams a fast, affordable way to find and fix critical vulnerabilities before attackers can exploit them, especially in large codebases.