Google Launches Gemini 3.8 Live Models with Real-Time Voice Intelligence and Multi-Step Reasoning

Google is expanding its Gemini 3.8 family with two new AI models designed to make conversational AI more intuitive and capable. The company announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, positioning them as production-ready voice assistants that enhance how users interact with AI across Google's ecosystem.1

1

2

2

Two Models Built for Different Needs

The two models serve distinct purposes within Google's AI strategy. Gemini 3.8 Live is built for scale and cost efficiency, combining conversational intelligence with fluid dialogue and real-time visual input processing. In contrast, Gemini 3.8 Live Extended Thinking targets high-complexity tasks requiring multi-step reasoning and increased intelligence.3

3

Both live dialogue models represent Google's push toward making voice agents more reliable and natural in their interactions.

Real-Time Processing and Language Detection

Gemini 3.8 Live processes visual inputs in near real-time while handling tools and API calls in the background. This allows the model to acknowledge requests and maintain conversation flow simultaneously. The model can detect and transition between 97 supported languages mid-conversation, making it adaptable for global users.1

1

This capability positions it as a practical solution for developers and enterprises seeking cost-effective voice agents that can scale across diverse user bases.

Leading Performance in Speech Quality and Task Completion

Gemini 3.8 Live Extended Thinking delivers what Google calls simultaneous reasoning and speaking, a feature that enables the model to think through complex problems while maintaining conversational flow. The model captured the top spot on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6.2

2

3

3

It also leads in agentic task completion with 68.6% on τ-Voice and achieved 35.1% on Sierra's τ-Voice-banking benchmark, which tests real-time, audio-native conversational AI agents on complex banking and fintech customer support tasks.1

1

The model scored 97.7% on Big Bench Audio, demonstrating strong reasoning capabilities while maintaining competitive pricing compared to other frontier models.2

2

Meanwhile, Gemini 3.8 Live secured second place in the Speech Agent Arena, showing high user preference alongside its cost-effectiveness.3

3

Integration Across Google's Ecosystem

Both models are rolling out across Google's product lineup. Gemini 3.8 Live will power Search Live in AI Mode, while Gemini 3.8 Live Extended Thinking is becoming available in Gemini Live.1

1

Users with AI Pro and Ultra subscriptions will find Live Extended Thinking in Google Workspace applications including Google Docs, Gmail, and Keep. The model supports recently launched features like Gmail Live for conversational search, Docs Live for draft generation and editing, and Keep Live for note creation.2

2

The model uses early verbal cues like "Let me check that..." to acknowledge prompts naturally and provides live progress narration to guide users through multi-step background tasks.2

2

This approach maintains conversational flow even during complex operations.

Developer and Enterprise Access

Developers can access both models through the Gemini API and Google AI Studio starting today.1

1

Enterprise users will find both models available in private preview through Gemini Enterprise, with plans to expand availability to Gemini Enterprise for Customer Experience and Google Workspace business customers. All audio generated by these models includes SynthID watermarking, Google's digital watermarking and detection tool that embeds invisible signals into AI-generated content.1

1

What This Means for Voice AI

The launch comes approximately two weeks after Google announced Gemini 3.8 Flash and follows the March release of Gemini 3.1 Flash Live.1

1

2

2

These new models demonstrate Google's commitment to near real-time reasoning and making voice interactions with AI feel more natural. For businesses, the combination of strong performance benchmarks and cost efficiency could accelerate adoption of voice agents in customer service and enterprise workflows. Watch for how these models perform in real-world banking and fintech applications, where the τ-Voice-banking benchmark results suggest promising capabilities for handling complex customer interactions.