Gemini 3.8 Live and Live Extended Thinking release

Google has released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two advanced voice dialogue models designed for natural, fluid conversations.

2 min read

Screenshot of Google AI Studio showing the Gemini 3.8 Live model selection menu.
Google AI Studio interface displaying Gemini 3.8 Live model selection options.

Google has introduced two advanced voice dialogue models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, designed to make voice interactions more natural, fluid, and intuitive.

Key facts

Top Speech Reasoning Model
StepAudio 3 Realtime (99.7%)
Top Conversational Dynamics Model
StepAudio 3 Realtime (98.9%)
Lowest Time to First Audio
Deepslate Opal (0.44s)
Most Affordable Input Audio
Qwen Audio 3.0 Realtime Plus ($0.0331/hour)
Top Speech to Speech Models by Reasoning Quality
StepAudio 3 Realtime
Model: StepAudio 3 Realtime
Big Bench Audio Score: 99.7%
Qwen Audio 3.0 Realtime Plus
Model: Qwen Audio 3.0 Realtime Plus
Big Bench Audio Score: 99.2%
Qwen3.5 Omni Plus Realtime
Model: Qwen3.5 Omni Plus Realtime
Big Bench Audio Score: 98.7%
Gemini 3.8 Live Extended Thinking (High)
Model: Gemini 3.8 Live Extended Thinking (High)
Big Bench Audio Score: 97.7%
Step-Audio R1.1 (Realtime)
Model: Step-Audio R1.1 (Realtime)
Big Bench Audio Score: 97.6%

Overview of the New Models

The standard Gemini 3.8 Live model is built for scale and cost efficiency, combining conversational intelligence with real-time visual grounding. It can process visual inputs in near real-time and automatically detect and switch between 97 supported languages mid-conversation.

Gemini 3.8 Live Extended Thinking is designed for high-complexity tasks, offering increased intelligence and multi-step reasoning while maintaining an uninterrupted conversational flow. It can reason and speak simultaneously, using early verbal cues to acknowledge prompts and live progress narration to walk users through background tasks.

Capabilities and Performance

Both models can execute tools and API calls in the background while keeping the conversation going. Gemini 3.8 Live Extended Thinking captures the top spot on Artificial Analysis' Speech to Speech Quality Index with a score of 82.6. It also leads in agentic task completion with 68.6% on the τ-Voice benchmark and 35.1% on Sierra's τ-Voice-banking benchmark, alongside scoring 97.7% on Big Bench Audio.

To help prevent misinformation, all audio generated by these products includes SynthID watermarking woven directly into the audio output.

Availability

Developers can access both models through the Gemini API and Google AI Studio, with support from developer platforms such as Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents. Enterprise users can find them in private preview in Gemini Enterprise, while general users can access them across Search Live, Google Workspace, and the Gemini app.

Source: Original announcement

Worth a read?

Report

Comments