Gemini Computer Use: what it does and what it costs
Gemini Computer Use is out of preview, with no free tier. Google's recommended model costs $0.75 per million input tokens until January 1, 2027.
本页内容
What is Gemini Computer Use?Which Gemini models support computer use?How much does Gemini Computer Use cost?Will a Gemini agent buy things on its own?What Gemini Computer Use means for youGemini Computer Use, the Gemini API tool that lets a model look at a screen and click and type for you, became generally available on October 30, 2026. It has no free tier, Google recommends Gemini 3.8 Flash for it rather than Gemini 4 Argon, and your own app has to run the browser and approve risky steps like payments.
What is Gemini Computer Use?
Gemini Computer Use is a tool in the Gemini API for building agents that control a browser, a phone or a desktop. Google's Computer Use documentation puts it this way: "Using screenshots, the model can 'see' a computer screen, and 'act' by generating specific UI actions like mouse clicks and keyboard inputs."
The model does not touch your computer itself. The same page says: "Similar to function calling, you will need to implement the client-side execution environment to receive and execute the Computer Use actions." So a developer's app takes a screenshot, sends it to Gemini, gets back an action such as a click, runs it, and repeats.
The Gemini API changelog entry for October 30, 2026 reads: "Released the Computer Use tool to general availability (GA) across the Interactions and GenerateContent APIs." Google DeepMind first announced Gemini 2.5 Computer Use as a public preview on October 7, 2025.
A common question is whether this works inside the Gemini consumer app. Google documents it as a Gemini API tool, which means it is something developers build into their own apps.
Which Gemini models support computer use?
Four models support Gemini Computer Use, according to the Model versions section of Google's documentation, read on November 2, 2026:
| Model | ID | How Google describes it |
|---|---|---|
| Gemini 4 Argon | gemini-4-argon | "Our most capable model, supporting computer use for complex reasoning and agentic UI workflows." |
| Gemini 3.8 Flash | gemini-3.8-flash | "The recommended model for computer use, featuring high-accuracy UI interaction and reliable tool calling." |
| Gemini 3.5 Flash-Lite | gemini-3.5-flash-lite | "A low-latency, cost-effective model supporting computer use." |
| Gemini 3 Flash Preview | gemini-3-flash-preview | "Preview model supporting computer use." |
Google calls Argon its most capable model, but its pick for most screen tasks is Gemini 3.8 Flash. The documentation lists browser, mobile and desktop environments.
How much does Gemini Computer Use cost?
Gemini Computer Use costs whatever the model you choose costs per token, and there is no free option. The Gemini Developer API pricing page, read on November 2, 2026, lists the Computer Use free tier as "Not available." For the paid tier it says: "Charged as regular tokens per model pricing (e.g., standard Gemini 3.8 Flash pricing)."
Gemini Developer API pricing page, paid tier, read November 2, 2026
Output tokens cost more. On the same page, per million output tokens: Gemini 3.5 Flash-Lite is $2.50, Gemini 3 Flash Preview $3.00, Gemini 3.8 Flash $3.75 and Gemini 4 Argon $10.00.
Two of those prices have end dates. Gemini 4 Argon goes from $2.00 to $4.00 input and from $10.00 to $20.00 output on Tuesday, December 1, 2026. Gemini 3.8 Flash goes from $0.75 to $1.50 input and from $3.75 to $7.50 output on Friday, January 1, 2027. If you are building something to run past those dates, budget for the later rates.
Will a Gemini agent buy things on its own?
Google built a confirmation step for that, and your app is the one that handles it. The Computer Use documentation lists seven built-in policy categories: FINANCIAL_TRANSACTIONS, SENSITIVE_DATA_MODIFICATION, COMMUNICATION_TOOL, ACCOUNT_CREATION, DATA_MODIFICATION, USER_CONSENT_MANAGEMENT and LEGAL_TERMS_AND_AGREEMENTS. A response can carry a safety_decision that marks an action as allowed, as require_confirmation, or as blocked.
When the decision is require_confirmation, Google's documentation says your app should get the user's approval before it runs the action. It also recommends running the agent in a sandbox (a virtual machine, a container or a separate browser profile) and sanitizing input.
People in forum threads also ask whether an agent will open a search engine instead of the site they named, and what happens at a CAPTCHA. Google's pages cover the confirmation and blocking rules above. They don't promise the agent will always pick the right page, so a test run on your own task is the way to find out.
What Gemini Computer Use means for you
Gemini Computer Use is a developer tool, so people who don't write code would use it through an app someone builds with it. If you do write code, you get a model that reads screenshots and proposes clicks, plus the job of building the sandbox and the confirmation step yourself. Start with Gemini 3.8 Flash, since Google recommends it, and price your tasks at the January 2027 rate.
How we checked: on November 2, 2026 we read Google's Gemini API changelog, the Computer Use documentation and the Gemini Developer API pricing page. We did not run an agent ourselves. For another recent Google launch explained in plain words, see our post on Gemini 4 Carbon.
To use Divux in your AI, add mcp.divux.si to any AI agent that supports MCP.