Google has recently launched the Gemini 3.5 Flash Cyber, particularly as a cheaper rival to large scale security models such as Anthropic and Open AI. It scans, finds and patches software vulnerabilities automatically.
Features and reason behind its creation
The model expands Google’s existing automated defence work, building on its internal code security agent, CodeMender. Gemini 3.5 Flash Cyber is currently limited to a restricted pilot programme, available only to government bodies and select enterprise partners through CodeMender. Google is going for a careful rollout.
Also Read | Can AI medical advice be dangerous? Lawsuit claims ChatGPT left man critically ill
Gemini 3.5 Flash Cyber’s lightweight design lets it run multiple fat, low-cost scans across large codebases instead of one expensive pass. CodeMender uses 3.5 Flash Cyber repeatedly across different sections of code to analyse vulnerabilities, then combines the results into a single security report.
Google says that because of being cheap and fast to run, the model can be built directly into regular software updates, daily code commits, and time-sensitive product launches without adding significant cost.
The model is running across several production systems of Google such as Chrome, Android, Google Cloud, Ads, and YouTube. After sharing results from the CyberGym benchmark, Google proved that running the new model up to five times per task through CodeMender, it delivered results competitive with much larger and costlier models.