Sunita Somvanshi
Google has released two related AI models called Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. Gemini 3.8 Flash is designed for long-horizon software engineering and complex enterprise workflows, while Gemini 3.8 Flash Cyber specializes specifically in finding and fixing security vulnerabilities.
Gemini 3.8 Flash handles what Google calls long-horizon software engineering, letting the model work through multiple steps in a coding project rather than just answering a single question. Google describes iterative tool calls and agentic workflows that allow the model to complete complex development tasks.
Flash Cyber specializes in security work. Developers can use it to find vulnerabilities in code and automatically generate validated patches. Google reports it reached 86.2 percent on CyberGym Pass@1 testing, while the company also says it achieved over 70 percent success on internal benchmarks measuring real-world vulnerability discovery across 20 programming languages.
The pricing makes this accessible for more developers. Google set introductory rates at 75 cents per million input tokens and three dollars and 75 cents per million output tokens through December 31, 2026. Google’s own documentation also acknowledges limitations such as hallucinations, ongoing jailbreak-resistance work, and occasional timeouts.
