Google is making its Gemini Tasks feature free for hundreds of millions of users across the United States, Canada, Japan, Korea, Australia, and India, powered by the newly released Gemini 4 Hydrogen model. The update, which rolls out globally on November 17, 2026, transforms the Gemini application from a conversational chatbot into an autonomous agent capable of executing multi-step workflows, including calling businesses, managing calendars, and processing purchases through Google Pay.
This release marks a strategic pivot for Google’s AI assistant ecosystem, moving beyond information retrieval into active task execution at consumer scale. By introducing a dedicated execution model optimized for speed and agentic workflows, Google aims to reduce token costs for developers while offering consumers a unified interface for daily automation. The rollout also introduces new payment and security protocols designed to handle sensitive, real-world interactions without requiring users to manually navigate fragmented interfaces.
Gemini 4 Hydrogen: Architecture and Performance
Gemini 4 Hydrogen is engineered as a high-efficiency execution model rather than a deep-reasoning frontier model. While Gemini 4 Carbon and Gemini 4 Argon handle complex architectural judgment and long-horizon reasoning, Hydrogen prioritizes rapid task completion across diverse agentic and coding workflows. The model operates across a 1 million-token context window and supports four distinct thinking levels: low, medium, high (default), and a new xhigh tier. This tiered approach allows users to scale latency and cost against reasoning depth, ensuring computational resources are allocated only when a task requires deeper analysis.
Performance metrics published by Google show Hydrogen surpassing Argon on the Artificial Analysis Intelligence Index at its default high thinking level, scoring 51 compared to Argon’s 50, and reaching 54 at the xhigh setting. When compared to its predecessor, Gemini 3.8 Flash, Hydrogen completes tasks 2.5 times faster at its high default setting. It also triples Terminal-Bench 4.1 scores, jumping from 18.2% to 57%, while cutting fabricated claims by more than 5x.
For developer workflows, Hydrogen demonstrates strong performance on codebase navigation and debugging. It outperforms similar-tier models on SWE-sweep by discovering and repairing bugs without breaking existing code. In early testing across complex graphs, SQL, and data-heavy enterprise workflows, partners reported Hydrogen running 2x to 2.8x faster while reducing token usage by 12% to 23% compared to 3.8 Flash. The model natively batches file-read and search calls in parallel rather than inspecting files sequentially, which significantly accelerates repository-wide analysis.
Illustrative image: blickpixel / Pixabay
| Metric | Gemini 4 Hydrogen | Gemini 3.8 Flash | Gemini 4 Argon |
|---|---|---|---|
| Artificial Analysis Intelligence Index (high) | 51 | N/A | 50 |
| Terminal-Bench 4.1 Score | 57% | 18.2% | N/A |
| Task Completion Speed (relative) | 2.5x faster | Baseline | N/A |
| Fabricated Claims Reduction | >5x cut | Baseline | N/A |
| Agent Arena Ranking | #5 | N/A | N/A |
Agentic Capabilities and Payment Integration
The Tasks toggle in the Gemini application enables the model to act as a personal assistant by executing multi-step instructions across connected applications. Users can forward emails to [email protected], text the assistant via RCS integration, or assign tasks directly through the application interface. The system connects with over 60 popular applications through the Connected Apps Platform, pulling context from Gmail, Google Drive, Airtable, and Dropbox to execute workflows like meal planning, expense tracking, and calendar management.
A significant expansion for United States users includes direct business interaction and commerce capabilities. The agent can call local businesses to compare pricing, check inventory, and book appointments. In a documented workflow, the model identified nearby dealerships, placed simultaneous calls, navigated hold music, and scheduled a service appointment based on calendar availability. For commerce, Gemini utilizes the Agent Payments Protocol (AP2) to process transactions via Google Pay. The system supports digital tokens on cards from American Express, Mastercard, and Visa, and allows users to set hard spending caps while retaining standard card rewards. All receipts route to Google Wallet, which is expanding to iOS alongside this release.
"We’re working with American Express (coming soon), Mastercard, and Visa to keep your purchases secure with digital tokens on supported cards."
To manage complex development pipelines, Google is introducing Orchestrator in Google Antigravity. This feature dynamically switches between Carbon and Hydrogen within a single session. On Terminal-Bench 4.0, Orchestrator achieved near-Carbon task success rates while consuming 24% less quota and completing workflows 19% faster. In parallel agent setups, Carbon coordinates while Hydrogen powers worker and verifier agents across isolated workspaces, enabling teams to ship complex projects faster without driving up token costs.
Security Architecture and Safety Guardrails
As Gemini transitions into autonomous task execution, Google has implemented a hardware-isolated execution environment. Each task runs in a dedicated virtual machine via Google Cloud’s Agent Compute Engine, sandboxed to prevent cross-user data leakage. The system employs prompt-injection filters that scan search results and webpages alongside Google Safe Browsing to restrict link access. Users retain control through configurable confirmation prompts; the agent pauses for approval before executing critical actions like purchases, though routine tasks can be set to “Always allow.”
Illustrative image: ann_zima / Pixabay
"This sandbox is just one of multiple layers of defense, including prompt-injection filters that scan search results, webpages, and more alongside Google Safe Browsing to limit which links Gemini can open."
Safety evaluations indicate substantial improvements over previous iterations. Hydrogen reduces overreach in agentic tasks by up to 81% and under-service by up to 79%. Misaligned actions and task shortcutting across tool-use and coding evaluations decreased by 12x. The model achieved a 94% honesty rate on internal agent honesty evaluations and demonstrated resilience against indirect prompt injection attacks, scoring 2.3x higher on the Gray Swan IPI benchmark. Chain-of-thought legibility remains at 99.9% across more than 200,000 audited turns, with zero deceptive reasoning edits.
Consumer-facing safeguards also address domain-specific accuracy. Updated filters distinguish between legitimate medical inquiries and biosecurity risks, keeping false refusal rates on HealthBench Pro under 0.4%. Google emphasizes that businesses can opt out of AI-initiated calls, and the system will not expose payment or private account information without explicit user authorization. The company has also opened an invitation to participate in the AI Vulnerability Reward Program (AI VRP) to stress-test agentic system security.
Developer and Enterprise Rollout
Gemini 4 Hydrogen is available across multiple Google surfaces starting November 17, 2026. Developers can access the model through the Gemini API in AI Studio and Android Studio, with pricing set at $1.50 per million input tokens and $7.50 per million output tokens across the full 1 million-token window. Enterprise customers can deploy the model via Gemini for business, Google Antigravity, and Gemini in Google Sheets for eligible subscriptions.
Consumer access extends beyond the Gemini application toggle. The model powers upgraded responses in Google Search, utilizing enhanced query fan-out techniques to surface broader link ranges and adaptive visual responses. Paid Google AI Pro and Ultra subscribers gain access to Hydrogen in Google Antigravity and Gemini in Google Sheets. The rollout also invites third-party developers to apply for the Connected Apps Platform, signaling a push toward standardized agentic integrations.
Google acknowledges that agentic web navigation remains an evolving standard. The company notes that some sites still require manual final-step interaction, and they are partnering with major websites and security providers to adapt web infrastructure for automated assistants. The release also coincides with an invitation to participate in the AI Vulnerability Reward Program (AI VRP) to stress-test agentic system security.
Sources
- blog.google: Put Gemini to work for you
- blog.google: Introducing Gemini 4 Hydrogen
- theverge.com: Google’s Spark AI agent becomes ‘Tasks’ — and now it’s free
- techcrunch.com: Google expands access to Gemini Tasks, which can help with calls, purchases | TechCrunch
PNEUMETRON EDITORIAL
Pneumetron's pipeline pairs AI-assisted drafting with human editorial review before publishing — our goal is to make staying informed easier, not to replace real reporting.




