Introducing computer use in Gemini 3.5 Flash

Computer use is now a built-in tool in Gemini 3.5 Flash to build agents that can interact across platforms.

Computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering our best performance yet for agentic computer use tasks. Previously only available as a standalone Gemini 2.5 computer use model, computer use is now integrated natively in the main Gemini Flash model. Gemini already excels at function calling and using built-in tools like Search and Maps grounding. With built-in computer use capability, developers can now use 3.5 Flash to reliably build custom agents that can see, reason and take action across browser, mobile and desktop environments. This unlocks improved performance for long-horizon and enterprise automation tasks like continuous software testing and knowledge work across professional applications.

Developers and enterprises can start using computer use in 3.5 Flash via the Gemini API and Gemini Enterprise Agent Platform.

Click Here

Open

Making computer use safe in 3.5 Flash

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. We’re also releasing two optional enterprise safeguard systems that enable enterprises to:

Require explicit user confirmation for sensitive or irreversible actions.
Automatically stop tasks if an indirect prompt injection is identified.

Taking a “defense-in-depth” approach, we encourage developers to combine these features with secure sandboxing, human-in-the-loop verification and strict access controls. Additional information on safety measures can be found in our best practices documentation.

We are already seeing customers drive value with computer use. Here’s what some of them have to say:

Quote from Magnus Muller, CEO, Browser Use

quote from Alvin Stanescu, Senior Director - UIPath

To start building with computer use today:

Try it now: Test the capabilities in a demo environment hosted by Browserbase.
Start building: Dive into our reference implementation and documentation via Gemini API and Gemini Enterprise Agent Platform.

Google Gives Gemini 3.5 Flash Computer Control Skills, AI Agents Can Now Click Buttons and Fill Forms

Google has officially integrated native “computer use” capabilities into Gemini 3.5 Flash, transitioning the feature from a standalone experimental model into a core, built-in tool. This upgrade allows the model to see, interpret, and autonomously interact with digital interfaces just like a human operator. [1, 2, 3]

By blending its large 1-million-token context window with intent-based visual actions, Gemini 3.5 Flash is highly optimized for complex, long-horizon automation workflows. [2, 4]

Watch these summaries and demonstrations of Gemini 3.5 Flash’s new capabilities:

· 1970 M01 1

Key Capabilities and Features

Cross-Platform Execution: Custom agents can interact seamlessly across browser, mobile, and desktop environments (ENVIRONMENT_BROWSER, ENVIRONMENT_MOBILE, ENVIRONMENT_DESKTOP). [5, 6]
Intent-Based Interaction: The model takes screenshots of the active environment, identifies interactive UI components (buttons, text boxes, sliders), and executes actions like clicking, typing, or scrolling. [2, 4]
Adjustable Thinking Levels: Developers can fine-tune the depth of the model’s reasoning loops to optimally balance processing quality, cost constraints, and execution latency. [2, 5]
Human-in-the-Loop (HITL): Supports highly customizable client-side functions, making it simple to hand over control to a human supervisor for manual intervention when necessary. [7]

Speed and Cost Benchmarks from Introducing computer

Google’s implementation strategy prioritizes high throughput and affordability for enterprise automation: [2, 4]

Benchmark Performance: Gemini 3.5 Flash scores 78.4 on the OSWorld-Verified leaderboard, placing it virtually neck-and-neck with premium frontier models like GPT-5.5 (78.7). [2]
Speed Superiority: Processing at roughly 289 tokens per second, it operates nearly four times faster than competing frontier models. [4]
Aggressive Pricing: Priced at $1.50 per million input tokens and $9 per million output tokens, it is less than half the operating cost of comparable industry alternatives. [4]

Integrated Safety and Enterprise Guardrails

Operating an operating system autonomously poses severe prompt-injection risks. Google has integrated targeted adversarial training alongside two optional enterprise safeguards: [2, 8]

Explicit Confirmation Prompts: Forces the agent to halt and request human validation before executing irreversible operations, such as completing financial purchases or deleting records. [2, 9]
Indirect Prompt Injection Filters: An automated monitoring layer that instantly terminates active workflows if it detects hidden malicious instructions embedded within web pages or document layouts. [2, 4]

How to Get Started

The feature is globally accessible through the Gemini API on Google AI Studio and the Gemini Enterprise Agent Platform. [2, 5]

Developers looking to experiment can explore Google’s official GitHub reference implementation or utilize the web-agent testing environment provided via partnership with Browserbase. The basic developer loop follows four key states: sending a task request → receiving the structured UI actions → executing the inputs via automated testing frameworks like Playwright → capturing and feeding back the new screenshot state. [2, 5, 10, 11]

If you are looking to build or integrate an agent, tell me:

What specific operating system or browser environment are you targeting?
What is the primary automation use case (e.g., continuous QA testing, form filling, data scraping)?

Introducing computer