Google has integrated computer use capabilities directly into Gemini 3.5 Flash, making the feature a native tool rather than a separate model. The move lets developers build agents that can see, reason about, and interact with software interfaces across desktop, browser, and mobile environments without switching between models.
“Computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering our best performance yet for agentic computer use tasks,” Mateo Quiros, product manager at Google DeepMind, wrote in the announcement. It was previously available only as a standalone Gemini 2.5 computer use model.
The model scored 78.4 percent on OSWorld-Verified, a benchmark that measures how well an AI can navigate real operating systems and applications, nearly level with GPT-5.5 at 78.7 percent and slightly above Claude Opus 4.7 at 78.0 percent, according to published comparisons. On the MCP Atlas multi-step workflow benchmark, Gemini 3.5 Flash led the comparison set at 83.6 percent, ahead of Claude Opus 4.7 (79.1 percent) and GPT-5.5 (75.3 percent).
Gemini 3.5 Flash already ships with a 1-million-token context window, function calling, and built-in tools such as Search and Maps grounding. With computer use added, agents can perceive on-screen elements via screenshots, identify UI components, and take actions, clicking, typing, scrolling, switching tabs, across applications.
Pricing is set at $1.50 per million input tokens and $9 per million output tokens, with caching options to reduce costs.
Enterprise adoption
Several companies are already integrating the capability. Salesforce is weaving Gemini 3.5 Flash into its Agentforce platform for multi-subagent enterprise tasks. Accounting software firm Xero is deploying it for autonomous workflows including supplier identification and tax form processing. UiPath, Browserbase, and Browser Use are also listed as early customers.
Safety measures
Google applied targeted adversarial training to reduce prompt injection risks for agents operating in live environments. The release also includes two optional enterprise safeguards: requiring explicit user confirmation for sensitive or irreversible actions, and automatically stopping tasks when indirect prompt injection is detected.
Google recommends a “defense-in-depth” approach combining these features with sandboxing, human-in-the-loop verification, and strict access controls.
Developers can start using the feature via the Gemini API and the Gemini Enterprise Agent Platform.
Sources: Introducing computer use in Gemini 3.5 Flash (Google Blog, June 24, 2026); Gemini 3.5 Flash benchmarks (MoodBook, June 2026)