Google has released Gemini 3.8 Flash, a new version of its fast AI model for programming, complex tasks, and autonomous agents. Along with it, the company has launched Gemini 3.8 Flash Cyber, which can analyze code, find vulnerabilities, and suggest fixes.
Google has released Gemini 3.8 Flash, a new version of its fast AI model for programming, complex tasks, and autonomous agents. Along with it, the company has launched Gemini 3.8 Flash Cyber, which can analyze code, find vulnerabilities, and suggest fixes.
The company announced this on the News from Google blog.
One of the main improvements in Gemini 3.8 Flash concerns tasks where the model needs to work independently for a long time. For example, when the AI assistant needs to independently understand the code, use several tools, check the result, find an error and try again.
So the new model can “work harder” on complex tasks. For example, take more reasoning steps and call tools multiple times. Because of this, it will sometimes spend more tokens. However, if saving tokens is important to the user, they can reduce the model’s effort level or stay on Gemini 3.7 Flash. In addition, the model’s price is the same as in the previous 3.7 Flash — $0.75 per million input tokens and $3.75 per million output tokens.
Social media users have also started comparing the price of the 3.8 Flash to other models. For example, user sergey.seda posted a graph showing that the model is similar in efficiency to the Claude Opus 5, but significantly cheaper.
View on Threads
In a Google demo, the model, at a single prompt, created a DOS version of Google Maps with location search, directions, and Street View. It also built an interactive 3D visualizer that shows the internals of various devices.
But Gemini 3.8 Flash Cyber is designed for cybersecurity. It can analyze large code bases, find vulnerabilities, and automatically generate patches. For example, in an internal company test, 3.8 Flash Cyber searched for various types of vulnerabilities in code in 20 programming languages and managed to find more than 70% of cases.
And in the CWE-Bench benchmark, where the AI has to fix the problem correctly, Gemini 3.8 Flash Cyber got 47.2% of successful patches on the first try. The company also claims that the Chrome team got 2.6 times more correct patches from Gemini 3.8 Flash Cyber than from the best large commercial models tested.
However, this model will not be made publicly available. Google is only making it available to verified organizations through the Fairwind Program. Potential users include government agencies, critical infrastructure operators, software developers, and cybersecurity teams.
The regular Gemini 3.8 Flash is now available via the Gemini API and Google AI Studio. Enterprise customers can use it in Gemini Enterprise, and Google AI Pro and Ultra users can use it in the Gemini app. The model has also appeared in AI Mode in Google Search.
Previously, dev.ua wrote that three weeks ago, Google released Gemini 3.7 Flash for programming and AI agents.
