Google unveils Gemini 3.5 — a model that automates complex tasks

Google introduced Gemini 3.5 at the I/O 2026 conference — a new version of its flagship model that blurs the line between understanding and action. Gemini 3.5 can execute complex multi-step tasks entirely automatically — through browsers, web applications, APIs, and enterprise systems. This isn't just a model that gives advice. It's an agent that acts.

What was previously unavailable

Previous models looked at a task, analyzed it, and told the person: here's what needs to be done. Gemini 3.5 works differently. It sees the screen the way you do. It clicks buttons, fills out forms, reads results, and makes decisions based on what it sees. This is a fully automated workflow. The scale of transformation is impressive: tasks that previously took weeks of manual work are now completed in hours or even minutes. The cost often turns out to be two times lower than when using competing models.

Speed and efficiency

3.5 Flash processes output tokens four times faster than other leading models on the market. But speed didn't come at the expense of quality. According to independent tests for complex coding and agent task execution, Gemini 3.5 Flash outperforms even Gemini 3.1 Pro — the previous top-tier version. This creates the ideal balance: fast execution means savings on API calls, and low cost means companies can afford to automate far more processes than before.

Real-world use cases

Large corporations are already moving Gemini 3.5 into production:

• Macquarie Bank — full automation of financial document processing, including contract analysis
• Shopify — demand forecasting and real-time customer behavior analysis
• Xero — processing tax forms, accounting operations, and document flow
• Databricks — analysis of large data volumes and system diagnostics

These aren't concepts. They're production solutions that process real money and documents.

Enhanced security

Google clearly understands: when an agent operates automatically, the risk is higher. The model has built-in safeguards — protection against cyber threats and CBRN risks (biochemical and nuclear weapons). Gemini 3.5 will refuse to perform dangerous or potentially harmful tasks, even if directly asked. This is critical for trust: companies must be confident that an automated agent won't cross dangerous boundaries.

Availability

3.5 Flash is already available in the Gemini app, in AI Mode in Google Search, and through the public Gemini API for developers. For corporations, there's the Gemini Enterprise Agent Platform with extended capabilities and support. The more powerful version, 3.5 Pro, is coming within a month.

What this means

The center of gravity in automation is shifting from human to machine. Startups and corporations have gained a tool that allows them to scale operations without hiring additional people — just add agents. For the end user, this takes browsers and applications to a new level: they stop being just an interface, they become assistants that act on your behalf. But remember the key point: an automated agent is a powerful tool that requires oversight. It can make mistakes. It needs to be checked.