Skip to main content
Back

Google Opens Gemini Tasks to Hundreds of Millions of Free Users

Google is unlocking agentic Tasks in Gemini for hundreds of millions of free users across six countries, powered by a new Gemini 4 Hydrogen model.

  • Google is making Tasks in Gemini free across six countries, powered by a new Gemini 4 Hydrogen model.
  • Agent can make phone calls, pay bills, book appointments, shop, and text over RCS.
  • Runs inside hardware-isolated VMs on Google Cloud's Agent Compute Engine, sandboxed per user.
  • Checkout uses Agent Payments Protocol (AP2) with Mastercard and Visa, Amex coming soon.
  • Supports 60+ connected apps; developers can apply to join the Connected Apps Platform.
  • Launches free in U.S., Japan, Korea, Canada, Australia, and India, advanced features U.S.-first.

Google is removing the paywall from Tasks in Gemini, which can make phone calls, inspect account records, pay bills, and complete multi-step errands. The consumer release runs on Google's new Gemini 4 Hydrogen model.

Rollout and access
Area Details
Free rollout Hundreds of millions of users across the U.S., Japan, Korea, Canada, Australia, and India
Advanced capabilities Launching first in the U.S.
Entry points The Task toggle in Gemini, RCS messaging, and email sent to gemini@google.com
Calling availability Limited to supported U.S. states at launch

From prompt to completed errand

Google's launch demonstrations show Gemini calling two car dealerships in parallel to schedule service, comparing an unexpected utility bill with historical PDFs in Gmail, and buying a USB-C charger with a discount code it found online. These workflows combine account data, browser navigation, phone calls, authentication, and payments within one task.

The machinery behind each task

  • Personal Intelligence: Gemini can use context from Gmail, Calendar, Photos, Drive, and payment cards stored with Google, avoiding a separate upload to another agent service.
  • Google Password Manager: Saved credentials allow the agent to sign in to third-party portals, including utility accounts, when a task requires private account data.
  • Agent Payments Protocol: AP2 uses digital tokens to represent payment credentials during checkout. Mastercard and Visa are supported at launch, with American Express planned. Existing card rewards and purchase protections still apply.
  • Connected Apps Platform: More than 60 services, including Instacart, can connect to Gemini. Google is accepting applications from developers that want to add integrations.
  • Messaging channels: Users can contact Gemini through RCS, the carrier messaging standard that supports richer media, and can include photos in an ongoing thread. Tasks can also arrive by email.
Gemini settings with Text and Email options under the Channels menu

A private VM for every account

Tasks run inside Google's Agent Compute Engine, which creates a hardware-isolated virtual machine dedicated to the user's account. Google says these environments are never shared between users and are regularly tested by its internal red teams.

Controls around agent execution
Control Purpose
Prompt-injection filters Inspect search results and webpages for malicious instructions intended to redirect the model
Safe Browsing Restricts which links and destinations the agent can open
Action confirmation Pauses before consequential steps such as clicking Buy; users can approve selected routine actions with Always Allow
Spending caps Set a hard financial limit before the task begins
Payment records Send receipts to Google Wallet, which is also expanding to iOS in the U.S.

Scheduled prompts gain hands and a wallet

Google's earlier scheduled actions feature allowed Pro and Ultra subscribers to run recurring prompts at set times, with a limit of ten active tasks. Tasks in Gemini adds web navigation, account access, calls, and payments, giving the assistant authority to carry a workflow through multiple external systems.

The release also places Google deeper into the agent competition with OpenAI and Anthropic. Gemini already reaches a large consumer audience and connects directly to Workspace data, saved passwords, payment cards, Android messaging, and Google Wallet. That distribution reduces the setup required to use an agent for everyday transactions.

Web friction still stops the agent

Many websites block automated traffic or use flows that agents cannot complete reliably. Gemini may navigate most of a process and then return control to the user for the final click. Google says it is working with major websites and security providers on open, verifiable standards for identifying and handling agent traffic.

Phone calling also has geographic limits. Within the U.S., the feature is unavailable in Indiana, Louisiana, Minnesota, Montana, and Nebraska at launch.

The release is consumer-facing, and Google's announcement leaves out a general-purpose Tasks API. Developers can currently pursue distribution through the Connected Apps Platform, which provides the integration route described for third-party services.

A reference design for agent authority

A per-account execution boundary limits the damage from a compromised task, content filters treat the web as untrusted input, confirmation gates reserve consequential decisions for the user, and spending caps bound financial exposure. The Agent Compute Engine, AP2 payment tokens, and Connected Apps Platform show how Google is dividing compute, authorization, payments, and integrations into separate layers that other agent systems can evaluate against their own designs.