Google has integrated its Duplex AI system into Gemini Tasks, allowing users to automate business phone calls. The feature leverages Gemini 4 Hydrogen and Neon models, marking a significant expansion of Duplex's capabilities.
TL;DR
- Google Duplex is now part of Gemini Tasks, enabling AI-driven business calls.
- The integration uses Gemini 4 Hydrogen and Neon models for enhanced functionality.
- This move positions Google's AI agents to handle complex, real-world tasks autonomously.
What happened
Google Research announced on November 20, 2026, that Duplex, its AI system for automated phone calls, is now integrated into Gemini Tasks. This integration allows Gemini to dial businesses on behalf of users when structured booking channels are unavailable.
The feature is powered by Gemini 4 Hydrogen and Neon models, which handle live calls and are part of broader Tasks workflows. These workflows include utility bill analysis, Agent Payments Protocol transactions, email, and RCS coordination.
Duplex has evolved through three generations: pre-LLM closed domains, generative hybrids with LaMDA, and prompted tasks on Gemini. The latest shift allows for generalized task handling without the need for vertical-specific training.
Why it matters
This integration positions Google's AI agents to handle more complex, real-world tasks autonomously. It is a direct response to competitors like Meta Muse, which also offers outbound calling capabilities.
For developers and startups, this move highlights Google's bet on prompt-defined missions over bespoke finetunes for each local service category. It also underscores the importance of latency and acoustic tuning for phone-based AI agents.
Investors should note that this integration is part of Google's broader strategy to consolidate its AI capabilities into a unified agent surface, potentially reducing the need for multiple specialized tools.
Key facts
- Duplex is now integrated into Gemini Tasks for autonomous phone errands.
- Gemini 4 Hydrogen and Neon models power the live calls.
- Calls open with a disclosure stating an automated Google system is in use.
- Businesses can opt out via Business Profile or verbal stop request during the call.
- Duplex has been used for trillions of business-info impressions in Search and Maps by 2023.
- Google lists pragmatic grounding, sub-second latency, and custom acoustic models as key optimizations.
- Duplex data populates Maps and Search listings, reducing repeat nuisance calls.
- Google did not publish API keys, pricing, or a public SDK in the Research article.
Context
Duplex was initially launched in 2018 and has since been integrated into various Google products like Search, Maps, and Pixel features. It has evolved from using recurrent networks to generative hybrids with LaMDA, and now to prompted tasks on Gemini.
This integration is part of Google's broader strategy to create a unified AI agent surface. It competes with other AI agents like Meta Muse, which offers paid outbound calling services in the US.
The move also reflects Google's focus on handling real-world tasks that require coordination between web forms, holds, and verbal exceptions. This is in line with the limitations discussed around other consumer agents like OpenAI Dots.
