Google Upgrades Gemini Task Mode with Gemini 4 Hydrogen and New E-Commerce Features
Google has updated Gemini's agentic Task mode with the faster Gemini 4 Hydrogen model, payment tokenization, and sandboxed security features.

Google has officially rolled out an upgrade to the agentic Task mode inside its Gemini assistant, powering the feature with a new, highly efficient AI model named Gemini 4 Hydrogen. Detailed by Josh Woodward, Vice President responsible for the Gemini app, the update aims to enhance how the assistant executes multi-step workflows such as scheduling appointments and comparing products online. Unlike Gemini's standard Chat mode, which generates quick responses to straightforward prompts, Task mode is engineered to perform complex, multi-stage actions while maintaining user control and security.
Speed, cost efficiency, and agentic workflows
The transition to Gemini 4 Hydrogen brings significant performance and architectural changes to Task mode. While Hydrogen is less powerful than Google's flagship Gemini 4 Carbon model, it offers a substantially higher level of cost efficiency and speed. Google reported that the new model can process certain tasks up to 2.8 times faster than its earlier entry-level predecessor. This acceleration is achieved in part by running project steps in parallel rather than processing them sequentially.
In addition to speed improvements, Gemini 4 Hydrogen introduces tighter human-in-the-loop controls. Users running Task mode can offer feedback after each step of a project to guide the agent's work. Before taking sensitive actions—such as completing an online purchase—the model proactively asks the user for explicit permission and instructions.
E-commerce integrations and isolated compute security
To bolster its e-commerce capabilities, Gemini's Task mode currently uses Google Wallet to handle transactions. Google disclosed that it is collaborating with financial networks American Express, Mastercard, and Visa to integrate payment tokenization technology. By substituting sensitive payment card numbers with placeholder token values during transactions, the technology minimizes the risk of data leaks to merchants. Furthermore, Google plans to implement the Universal Commerce Protocol, or UCP, to standardize how AI agents interact with digital storefronts, helping reduce transaction errors.
Data security during agentic execution relies on Google Cloud's infrastructure. Task mode can draw upon context from other Google tools, such as Gmail purchase histories, without transferring data off native systems. Google processes these workloads inside the Agent Compute Engine, executing tasks in hardware-isolated, account-dedicated virtual machines. Each sandboxed environment undergoes routine testing by internal security red teams to ensure tenant isolation.
What it means for developers
Google's decision to offer Gemini's Task mode for free across the United States, Japan, Korea, Canada, Australia, and India marks a sharp contrast to competitors. Both OpenAI Group PBC and Anthropic PBC currently charge for comparable multi-step agentic capabilities within their respective chatbot services. This free tier from Google could prompt competing providers to revise their pricing models.
For software engineering teams and AI builders evaluating agentic frameworks, Google's architectural blueprint demonstrates how lighter, specialized models like Gemini 4 Hydrogen can execute structured multi-step tasks at a fraction of the latency and compute cost of massive flagship models. As multi-model strategies become standard across enterprise applications, developers can try top AI models cheaply through one API at https://apixoai.online to evaluate performance across different workloads.
Additionally, Google’s deployment strategy highlights the growing necessity of secure sandboxing through isolated virtual machines and open standards like payment tokenization and UCP when moving autonomous AI agents into real-world production environments.
Source: Google upgrades Gemini’s agentic Task mode with new model, e-commerce features — SiliconANGLE AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key

