The main addition is Auto Browse — a feature that lets Gemini perform multi-step actions on websites instead of only answering questions about a page.
What Gemini in Chrome can do
Gemini can:
- summarize long articles;
- answer questions about the current webpage;
- connect with Google services such as Keep and Calendar;
- generate and edit images with Nano Banana;
- use websites on the user's behalf through Auto Browse.
Auto Browse can handle tasks such as:
- filling out forms;
- ordering groceries;
- booking parking;
- navigating multi-step website flows.
For sensitive actions, Gemini asks for confirmation before proceeding.
Technical requirements
| Parameter | Requirement |
| Platform | Android |
| Android version | Android 12 or newer |
| RAM | At least 4 GB |
| Browser | Latest Chrome version |
| Account | Google account required |
| Language | English |
| Incognito | Not supported |
| Auto Browse subscription | Google AI Pro or AI Ultra |
| Region | India |
Auto Browse is available on both desktop and Android.
Auto Browse limits
Google currently applies daily limits:
- Google AI Pro: up to 20 multi-step requests per day;
- Google AI Ultra: up to 200 multi-step requests per day.
This indicates that Auto Browse uses more compute than a regular Gemini query because one task can require multiple model calls and browser actions.
Previous version vs new Android release
| Previous implementation | New Android version |
| Gemini mainly answered questions | Gemini can also perform actions |
| User manually interacted with websites | Auto Browse can click, select and fill forms |
| Limited browser automation | Multi-step browser workflows |
| Mostly assistant-style interaction | Agent-style interaction |
| Desktop rollout came first | Now available on Android |
The key change is not the chatbot interface itself, but browser control.
Previously:
Page → Gemini → Answer
With Auto Browse:
Page → Gemini → Action → New page state → Next action
Security model
Browser automation creates additional risks because Gemini can interact with authenticated websites.
Google uses several safeguards:
- confirmation before sensitive actions;
- permissions for browser control;
- Safe Browsing requirements;
- protection against prompt-injection attacks.
Prompt injection is especially relevant for browser agents because malicious instructions can be embedded directly inside a webpage.
What changed technically
The Android release does not represent a major change in the underlying Gemini model.
The main change is expanded access to browser tools:
- page context;
- UI elements;
- navigation;
- form inputs;
- multi-step actions;
- Google service integrations.
The shift for software engineers is a change from an assistant that works in a browser to an agent that acts in a browser.
Gemini no longer only explains what to do on a website. Auto Browse can execute the workflow itself.
