Google integrates computer use natively in Gemini 3.5 Flash for browser and desktop automation

Google on June 24 announced that computer use — the ability for an AI model to see, reason about, and take action across digital interfaces — is now a built-in capability in Gemini 3.5 Flash. Previously available only as a standalone Gemini 2.5 computer-use model, the feature is now integrated directly into Gemini 3.5 Flash, Google's primary developer-facing multimodal model. The company said it delivers its "best performance yet for agentic computer use tasks."

What's new

Computer use in Gemini 3.5 Flash allows developers to build custom agents that operate across browser, mobile, and desktop environments using a single model rather than a separate endpoint. Google says the integration "unlocks improved performance for long-horizon and enterprise automation tasks like continuous software testing and knowledge work across professional applications."

Key capabilities:

• Visual perception and action: The model analyzes interfaces and executes commands across digital environments without a text-to-action intermediary
• Long-horizon task support: Enables continuous software testing and enterprise automation workflows
• Native toolchain integration: Works alongside Gemini's existing function calling and built-in tools including Search and Maps grounding

Access is available now through the Gemini API and the Gemini Enterprise Agent Platform.

Context

Computer use as a deployable API capability became prominent when Anthropic added it to Claude in late 2024. Since then, every major frontier lab has moved to offer similar capabilities. Google initially offered computer use through a dedicated Gemini 2.5 model — a standalone endpoint that developers had to manage separately from their general-purpose API usage.

Gemini 3.5 Flash launched earlier in 2026 as Google's high-throughput developer model in the 3.5 family, designed for scale and cost efficiency. Folding computer use into this model consolidates what previously required two separate APIs into one.

Why it matters

Bringing computer use into a general-purpose model rather than isolating it in a specialty endpoint represents a meaningful architectural shift. Developers building on Gemini 3.5 Flash can now add browser and desktop automation to existing applications without switching models or managing additional configurations.

For enterprise use cases — automated QA, document processing across professional tools, multi-step browser workflows — this integration lowers both the integration cost and the context-switching overhead. Google has accompanied the launch with a defense-in-depth safety architecture: targeted adversarial training specifically for computer use tasks, optional enterprise safeguard systems requiring explicit user confirmation before sensitive or irreversible actions, and automatic task termination when indirect prompt injection is detected.

That last point addresses one of the more serious risks in agentic computer use deployments, where a malicious page could attempt to redirect an agent mid-task. Having the model trained to recognize and halt on injection attempts — rather than relying purely on application-layer filtering — is a meaningful addition for production deployments.