Google has introduced Gemini 3.8 Flash, the latest iteration in its series of lightweight multimodal models. This release continues the company's three-week update cadence, following the debut of Gemini 3.7 Flash in mid-August. The new model is designed to function as an intelligent workhorse, with specific optimizations for software engineering, agentic workflows, and complex, multi-step reasoning tasks.
The model is currently available to Google AI Pro and Ultra subscribers through the Google Gemini app, AI Mode in Google Search, and Gemini in Google Sheets. For developers and enterprise users, the pricing remains consistent with the previous version, set at $0.75 per million input tokens and $3.75 per million output tokens. Google notes that the model achieves performance gains by executing additional reasoning steps and iterative tool calls, though users can opt for lower effort levels if compute efficiency is the primary requirement.
Alongside the standard release, Google debuted Gemini 3.8 Flash Cyber, a variant specifically tuned for vulnerability detection and automated patching. This model is not available for general public use; instead, it is being distributed through the newly launched Fairwind Program. Access is restricted to vetted entities, including government agencies, national cyber authorities, critical infrastructure operators, and core technology platforms.
Google claims that Gemini 3.8 Flash Cyber demonstrates frontier-level performance in autonomous vulnerability discovery. Internal testing conducted by the Chrome Security team reportedly showed that the model produced 2.6 times more correct patches for Chrome vulnerabilities than larger commercial alternatives. Additionally, the model achieved a 47.2% pass rate on the CWE-Bench patching benchmark, positioning it near the performance of leading frontier models at a lower cost.
The underlying architecture for both models relies on long-running agentic loops that recursively evaluate and refine outputs. While the standard Flash model is built for general long-horizon coding and autonomous tasks, the Cyber variant focuses on defensive security capabilities. Google stated that it has prioritized vulnerability remediation over offensive exploitation features in the development of the Cyber model to better support security researchers and defenders.