Introducing computer use in Gemini 3.5 Flash

Details

Google DeepMind has embedded computer use as a built-in tool in Gemini 3.5 Flash, consolidating a capability that previously required a separate Gemini 2.5 computer use model. Developers building agents on Flash can now add browser, mobile, and desktop automation without switching to a different model endpoint. The announcement targets long-horizon enterprise tasks: continuous software testing, knowledge work across professional applications, and complex multi-step workflows. Flash is positioned as the cost-efficient tier for production computer-use workloads, offering improved performance over the prior standalone version while being more economical than the Pro model. On safety, Google is shipping two enterprise safeguard systems: explicit user confirmation requirements for sensitive or irreversible actions, and automatic task termination when indirect prompt injection is detected. The company used targeted adversarial training to reduce prompt injection risks for agents operating in live environments. Best practices documentation recommends combining these with secure sandboxing, human-in-the-loop verification, and strict access controls. Access is available now via the Gemini API and Gemini Enterprise Agent Platform. Browserbase hosts a demo environment. This is a shipping product, not a research preview — enterprise teams evaluating computer-use agent infrastructure can start building today.