Google launches Gemini 3.8 Flash models for software and cybersecurity

Google unveiled two new versions of its 3.8 Flash models on Wednesday, targeting both software development and cybersecurity. The standard Flash model is designed for agentic tasks, while the Flash Cyber variant focuses on detecting and mitigating vulnerabilities. This announcement marks another step in Google’s ongoing efforts to enhance AI capabilities across different fields.

The release comes at a time when the demand for advanced AI tools is surging, particularly in software engineering and cybersecurity. Google CEO Sundar Pichai emphasized that the 3.8 Flash models deliver “significant leaps” in performance compared to their predecessor, 3.7 Flash. The new Flash model has already outperformed many leading models on the DeepSWE coding benchmark, showcasing its efficiency and effectiveness at a lower cost.

Flash Cyber has been described as Google’s “most capable” cybersecurity model, achieving frontier-level performance in vulnerability discovery and patching. It scored 86.2% on the CyberGym cybersecurity benchmark and 47.2% on the CWE-Bench, which assesses AI patching capabilities. In an internal evaluation, Flash Cyber demonstrated a success rate exceeding 70%, further solidifying its role in enhancing cybersecurity measures.

As these models roll out, they are likely to impact various industries. Organizations may find new ways to streamline software development and bolster their cybersecurity efforts. The advancements in Google’s Flash models underscore the growing importance of AI in addressing complex challenges in technology, making it a critical area to watch in the coming months.