Google recently launched Gemini 3.5 Flash, which focuses on "computer use" capabilities, which makes models better at directly manipulating computer interfaces, performing multi-step tasks, and acting as more active agents in coding, research, and complex workflows. This update means that Gemini's positioning has further shifted from a traditional question and answer tool to an executive AI that can actually "do it for you".
Gemini 3.5 Flash is described by Google as one of its strongest models to date for coding and autonomous agent tasks. Officials emphasize that it combines low latency with high quality, and performs better than 3.1 Pro in multiple benchmark tests. Google also said that this version has significantly improved the output speed and is suitable for scenarios that require quick feedback and continuous operations, such as programming, tool calling, and automated office work.
In terms of practical capabilities, Gemini 3.5 Flash can better handle long-chain tasks, such as independently executing coding processes, managing research projects, and even participating in the complex work of building an operating system from scratch in internal testing. Google said that this model is not only "faster", but also improves both quality and speed through optimized Flash architecture, making it more suitable for AI agent scenarios that require continuous interaction.
Google has also embedded this capability into its own products and platforms. The AI mode for Gemini apps and search already uses Gemini 3.5 Flash by default, and is also available through the Gemini API, Gemini Enterprise, and related developer tools. Google also mentioned that search-oriented agent functions and a new generation of personal AI agents Gemini Spark will also be supported by this model, with the goal of allowing users to complete daily digital tasks with less human intervention.
In terms of security, Google said that with the popularization of stronger autonomous agent capabilities, risk control has also been simultaneously strengthened, especially network security and CBRN-related security protection. When this type of model encounters sensitive issues, it does not simply refuse to answer, but prefers to process the request within a controllable framework and pause to wait for user confirmation when necessary. Overall, Gemini 3.5 Flash represents an important advancement by Google in the direction of “letting AI perform tasks directly”.