Muitto
Voltar ao catálogo
Texto
gemini
Equilibrado

Gemini 2.5 Computer Use Preview 10-2025

Gemini 2.5 Computer Use Preview 10-2025 automates desktop tasks through natural language.

A tradução deste card ainda não está disponível; parte do conteúdo aparece em inglês.

Janela de contexto (tokens)
131.072
Sobre este modelo

Gemini 2.5 Computer Use Preview 10-2025 is an advanced text model built to interact with graphical user interfaces the way a human would. It reads on-screen elements, understands application layouts, and carries out sequences of actions to achieve a goal described in natural language. Whether filling out forms, organizing files, or navigating complex software, it bridges the gap between intent and execution.

Use this model when you need to automate repetitive desktop workflows, guide users through multi-step software processes, or perform tasks that require understanding visual context alongside text input. It excels at scenarios where traditional scripting falls short because the interface changes or requires contextual judgment.

What makes Gemini 2.5 Computer Use Preview 10-2025 distinctive is its ability to reason about screen state, plan a sequence of actions, and adapt when interfaces don't behave as expected. It combines language understanding with interface awareness, making it possible to automate tasks that normally require a human at the keyboard.

Melhores casos de uso
  • Automating repetitive form filling and data entry
  • Navigating multi-step software workflows
  • Organizing and renaming files in bulk
  • Guiding users through unfamiliar applications
  • Executing cross-application copy-paste workflows
Pontos fortes
Interface Awareness

Understands graphical interfaces and can identify interactive elements like buttons, fields, and menus.

Multi-Step Planning

Breaks down complex tasks into logical action sequences and executes them in order.

Adaptive Execution

Adjusts its approach when interfaces change or unexpected dialogs appear mid-task.

Dicas
  • Describe tasks in clear sequential steps rather than vague high-level instructions for best results.
  • Break very long workflows into smaller sub-tasks to improve reliability and reduce error accumulation.
  • Specify exact application names and menu paths when the task could be interpreted multiple ways.
  • Review the model's planned actions before execution on sensitive or irreversible workflows.
Limitações
  • May struggle with heavily customized or non-standard interface designs that deviate from common patterns.
  • Performance degrades with tasks requiring precise pixel-level interactions or fast real-time responsiveness.
  • Cannot access or interact with content behind authentication barriers or locked screens without explicit user setup.
Exemplos de geração
Exemplo em breve
Cross-app data transfer

Open the CRM application, copy all customer contact details from the current view, switch to the email client, and paste them into a new contact group called VIP Clients.

Exemplo em breve
Automated research compilation

Open the browser, search for the top five competitors for electric scooters, copy each company name and founding year into a new text document, then save it as Competitor Analysis.

Exemplo em breve
Multi-step form submission

Open the procurement portal, fill in the vendor registration form using the details from the file called Vendor Info on the Desktop, upload the certificate in the attachment field, and submit the form.

Prompts de exemplo
Fill out a spreadsheet
Open the spreadsheet called Q3 Sales, fill column D with the product category for each item based on its name in column B, then save the file.
Organize downloads folder
Go to the Downloads folder, create subfolders named Documents, Images, and Installers, then move each file into the matching folder based on its file type.
Export a report
Open the accounting app, navigate to the Reports section, generate a monthly summary for October 2025, and export it as a PDF to the Desktop.
Parâmetros
seed

Provides a starting point for generation so the same input produces more consistent, repeatable outputs.

stop

Specifies words or phrases where the model should stop generating, useful for controlling output boundaries.

top_p

Limits the model's word choices to the most likely options, making outputs more focused and reliable.

max_tokens

Sets the maximum length of the model's response in tokens, which roughly correspond to word fragments.

temperature

Controls how creative or predictable the model's responses are; lower values keep it focused and deterministic.

Preços
—

Os preços exibidos são o que você paga na Muitto. Sem surpresas.

Perguntas frequentes

Can this model control any application on my computer?

It can interact with most standard graphical applications, but performance varies depending on interface complexity and customization.

Does it work with keyboard shortcuts and mouse actions?

Yes, it can simulate both keyboard input and mouse actions including clicks, drags, and scrolls.

Is it safe to let the model perform actions automatically?

Always review planned actions before execution, especially for tasks involving file deletion, financial transactions, or system settings changes.