Skip to content
Theme
OnDevice Tools

AIAI Tools & Apps

Gemini tasks go free in six countries with new Hydrogen model

Google has opened the Gemini tasks feature to free users in the US, Japan, Korea, Canada, Australia and India. The agent runs in a sandboxed virtual machine and is powered by a new model called Hydrogen.

Illustration of a phone showing the Gemini app with Chat and Task options above the prompt box

Key takeaways

  • Gemini tasks are now free in the US, Japan, Korea, Canada, Australia and India.
  • The feature was previously called Spark and required an AI Pro subscription.
  • Tasks run in a sandboxed virtual machine with protections against prompt injection.
  • Google says the feature is powered by a new model named Hydrogen.
In this article
  1. What changed in the Gemini app
  2. How the agent is built and kept in check
  3. The examples Google gives
  4. Why Google is pushing personal context

What changed in the Gemini app

Google has made the tasks feature in its Gemini app available to free users, according to 9to5Google. The capability was previously limited to AI Pro subscribers and was known as Spark before it was folded into the main Gemini app in October.

On mobile, a Chat and Task toggle now sits above the prompt box, while on the web it appears below. Chat opens the familiar chatbot, and Task lets people hand off longer jobs to Gemini. The free rollout covers the US, Japan, Korea, Canada, Australia and India.

How the agent is built and kept in check

Illustration: AI, Google logo, settings, Gemini logo and the labels AI Pro, Gemini, Google

Google says tasks run inside a hardware-isolated virtual machine, a separate computing environment that is sandboxed from the rest of the system. The company says this setup includes protections against prompt injection, a type of attack where hidden instructions try to steer an AI agent. When the agent browses the web, it uses Google Safe Browsing.

Other safeguards include asking for confirmation before a purchase and posing clarifying questions when an instruction is unclear. The feature is powered by a new model that Google calls Hydrogen, according to 9to5Google.

The examples Google gives

Google frames the goal as saving people time and money. In one example shared with 9to5Google, a user gave Gemini access to a utility portal through the new Google Password Manager integration and asked it to review past bills in email. The agent identified a heat pump running during peak pricing as the likely cause of a high bill.

In another example, a user asked for help choosing a USB-C charger for a trip. Google says Gemini compared options through its Shopping Graph, asked follow-up questions about wattage and colour, and helped complete the purchase through the new Google Pay integration. These are company-provided anecdotes, not independent test results.

Why Google is pushing personal context

Google argues that Gemini is a strong personal assistant because email, calendar, passwords, photos, files and payment cards already sit in its services. The pitch is that people do not need to move that data to another provider.

That argument also explains the caution around the feature. Giving an agent access to bills, messages and payment details raises privacy and security questions, which is why the sandboxing and purchase confirmation steps matter. Readers who want to try tasks should review which Google integrations they enable and check Google's own support pages for current availability and limits.

Sources

  1. You can now assign Gemini ‘tasks’ for free in six countries9to5Google

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.