Gemini 3.8 Live runs background APIs without pausing speech

In an official announcement published by Google DeepMind, the company introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The release focuses on solving one of the main bottlenecks in voice-driven AI: latency caused when a system needs to pause speaking to trigger an external database query or API call.

How do background APIs change daily voice workflows?

Traditional voice assistants force a rigid trade-off. When an AI agent needs to check inventory, execute a booking, or query a CRM, it either freezes the audio pipeline or disconnects the conversation loop while processing. According to Google DeepMind, Gemini 3.8 Live eliminates this pause by executing tools and calling APIs in the background while maintaining open audio streaming.

In operational terms, the system uses natural verbal cues—such as acknowledging a request with a brief filler phrase—while executing multi-step backend functions simultaneously. For customer service teams and internal helpdesks, this transforms voice interactions from rigid command-and-response scripts into fluid conversations where background system lookups occur alongside active speech.

What do the benchmark numbers mean in practice?

Google reports that Gemini 3.8 Live Extended Thinking achieved a 68.6% success rate on the τ-Voice benchmark for agentic tasks, along with 35.1% on Sierra’s τ-Voice-banking benchmark and top ranking on the Speech to Speech Quality Index with an 82.6 score.

Evaluating these metrics requires context. A 68.6% score on general agentic tasks means the model completes approximately seven out of ten multi-step voice operations successfully without human intervention or lost state. In strictly regulated environments like banking, the 35.1% metric reflects how complex authorization rules and multi-layered security protocols still challenge full end-to-end voice automation. Furthermore, built-in automatic language switching across 97 languages allows international support operations to handle multi-lingual callers on a single infrastructure stack without explicit manual routing.

What changes for Workspace users and developers on Monday?

For enterprise teams using Google Workspace, the introduction of live audio capabilities across Docs, Gmail, and Keep means document creation and inbox management can happen verbally while browsing other tools. Instead of issuing a voice prompt and waiting for generation, users can refine, redirect, or add context mid-output without resetting the session.

For developers building on the Gemini Live API, integration with platforms like LiveKit, LangChain, and Vercel offloads media streaming management. This allows technical teams to focus on function calling logic rather than managing custom audio buffering. All generated audio includes SynthID watermarking, establishing a technical audit trail for enterprise compliance.