Google Introduces Gemini 3.8 Flash and Flash Cyber for Agentic Workflows and Cybersecurity

Published: September 3, 2026 – 08:46 UTC

Gemini 3.8 Flash has become Google’s new model for reasoning, coding, and agentic workflows.
A separate Flash Cyber version is focused on vulnerability discovery and automated code fixes, and is available to vetted security professionals.

Google unveiled two new artificial intelligence models — Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. They are built on a single foundational core, but are designed for different use cases.

Gemini 3.8 Flash is positioned as a model for reasoning, coding, and executing agentic tasks. It is available at the same starting price as Gemini 3.7 Flash — $0.75 per million input tokens and $3.75 per million output tokens.

Gemini 3.8 Flash Cyber is geared toward cybersecurity. It is available to a limited group of vetted security professionals through the Fairwind program.

Gemini 3.8 Flash Bets on Autonomy

According to Google, Gemini 3.8 Flash shows significant progress over the previous version in software development, agentic tasks, and multi-step analysis.

In the DeepSWE v1.1 benchmark, the model outperformed most larger frontier models in its ability to autonomously complete complex engineering tasks end to end. Gemini 3.8 Flash also delivered better results than Gemini 3.7 Flash and other leading models on professional and quantitative tasks.

Google attributes the performance gains to an architectural approach in which the model takes more reasoning steps and more frequently uses tools iteratively. At higher compute effort levels, it can use more tokens to improve output quality.

Developers for whom compute efficiency matters more can use lower effort levels or continue working with Gemini 3.7 Flash, which remains supported.

Gemini 3.8 Flash Cyber Is Aimed at Defenders

Gemini 3.8 Flash Cyber is available through the Fairwind program to a limited group of vetted cybersecurity professionals.

In the CyberGym benchmark, the model demonstrated state-of-the-art autonomous vulnerability discovery and outperformed Gemini 3.5 Flash Cyber as well as significantly larger frontier models.

Google also tested the model on an internal benchmark covering complex codebases across 20 programming languages. The successful vulnerability detection rate exceeded 70%.

Special attention during development was given to automatic vulnerability remediation. Google noted that the priority for Gemini 3.8 Flash Cyber was defensive capabilities, not exploit creation.

The Model Is Already Being Used to Protect Google

Google said that Gemini 3.8 Flash Cyber is already being used to protect its internal codebase.

With the help of the model, the Chrome security team produced 2.6 times more correct vulnerability patches than with the best, much larger commercial models.

During internal testing, Wiz recorded 7.5-9.7% higher vulnerability detection recall compared with other leading models. At the same time, the cost of using Gemini 3.8 Flash Cyber was 2.3–5.2 times lower.

In addition, the Google Cloud team responsible for vulnerability research used the model to identify a critical, fundamental vulnerability in less than two hours. According to the company, finding and investigating such issues typically takes months.

Сообщение Google Introduces Gemini 3.8 Flash and Flash Cyber for Agentic Workflows and Cybersecurity появились сначала на INCRYPTED.

Source: Incrypted