Google has introduced two new artificial intelligence models. The systems have been named Gemini 3.8 Flash and Flash Cyber. Both models are built on a single foundational computational core. However, the systems are designed for entirely different use cases. The first model is oriented toward reasoning and code programming. The second version was created for modern cybersecurity tasks. The cost of use has remained at the same starting level. One million input tokens costs seventy-five US cents. One million output tokens costs three dollars and seventy-five cents. This price is identical to the cost of the Flash 3.7 version.
The Flash model demonstrates substantial progress in software development. The system handles agentic and multi-step analysis tasks with excellence. In the DeepSWE v1.1 benchmark, the model outperformed most major competitors. The system autonomously completes complex engineering tasks from start to finish. The model showed the best results in professional and quantitative tasks. The performance gains are explained by a special architectural approach taken by the company's developers. The model performs more reasoning steps and iteratively uses tools. Developers can choose lower levels of computational effort. Gemini 3.7 Flash remains a fully supported system at this time.
The Cyber version was created for vetted cybersecurity professionals. Access is provided through the specialized Fairwind program to a limited group. In the CyberGym benchmark, the model demonstrated a leading level of vulnerability discovery. The system outperformed the 3.5 Flash Cyber version and major competitors. The internal test covers complex codebases spanning twenty programming languages. The rate of successful vulnerability identification exceeded seventy percent. Developers made defensive capabilities the priority of the new model. Exploit creation was not the primary goal of this development.
The model is already being used to protect Google's internal codebase. The Chrome security team received 2.6 times more correct patches. Wiz recorded a higher completeness of vulnerability detection in the system. The cost of use was several times lower than that of competitors. The Google Cloud team identified a critical fundamental vulnerability very quickly. Finding such issues typically takes months of work by experts. The new model found the vulnerability in under two hours.