According to Google DeepMind, Gemini 3.5 Flash can now support agents that can observe, reason about and interact with browser, mobile and desktop environments. The company says the capability is designed for building automation systems that can carry out extended workflows across software applications, including tasks such as continuous software testing and professional knowledge work.
“Computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering our best performance yet for agentic computer use tasks,” said Mateo Quiros, Product Manager at Google DeepMind.
The move consolidates capabilities that previously required a separate computer use model. Developers can now access computer use and other built-in tools from the same model context rather than managing separate systems.
Google highlighted demonstrations showing Gemini 3.5 Flash using computer use to analyze the Gemini app and generate a categorized list of features. The company also showcased the model auditing documentation for accessibility issues.
The launch includes a focus on security controls for agents operating in live software environments. Google said it applies targeted adversarial training to help reduce prompt injection risks associated with computer use tasks.
“To mitigate some of the prompt injection risks for agents operating in live environments, we use targeted adversarial training for computer use in Gemini 3.5 Flash,” Quiros said.
Alongside the model-level protections, Google is introducing two optional enterprise safeguard systems. One requires explicit user approval before an agent performs sensitive or irreversible actions. The second can automatically stop a task when an indirect prompt injection attempt is detected.
The company said these protections are intended to be part of a broader security strategy and recommends that developers combine them with sandboxing, human oversight and strict access controls when deploying agents in production environments.
Developers can begin using the feature through the Gemini API, while enterprise customers can access it through the Gemini Enterprise Agent Platform. Google also noted that teams can evaluate the capability in a hosted browser environment provided by Browserbase before deploying it within their own infrastructure.
While Google describes Gemini 3.5 Flash as delivering its strongest performance so far for agentic computer use, the company did not publish benchmark comparisons against the standalone Gemini 2.5 computer use model or competing systems. Pricing details for large-scale enterprise deployments were also not disclosed.
This analysis is based on reporting from Google.
Image courtesy of Cole Kan/PCMag/Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
