Robots Atlas>ROBOTS ATLAS
GL
AI Models›Gemini

Gemini 3.8 Live

3.8 Live · Family: Gemini
Google's cost-efficient voice model — near real-time conversation and visual grounding, 97 languages, SynthID audio watermarking.
✓ Active✓ Public accessMultimodalAudioAudio📁 Gemini
Release date
15 September 2026
🏢Google DeepMindProducer🏢GoogleOwner
Access:APIHostedDeployment:☁ Cloud

Overview

Gemini 3.8 Live is Google DeepMind's cost-efficient voice model, combining conversational intelligence, fluid dialogue and visual grounding, built for scale. It processes voice and visual inputs in near real time, supports 97 languages with automatic mid-conversation language detection and switching, can call tools and APIs, and executes background tasks without interrupting the dialogue. All generated audio is watermarked with SynthID.

The model rolled out on September 15, 2026 — for developers via the Gemini API and Google AI Studio, and for users in Search Live, among others. It is marketed as highly cost-effective and ranked second in the Speech Agent Arena. A higher-tier variant, Gemini 3.8 Live Extended Thinking, targets high-complexity tasks.

Classification
MultimodalAudioAudio
Family: Gemini
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
✓ Tools
📥 Input: text, audio, image, video

Technical specification

License
Proprietary
Features:✓ Tool use
Modalities
⬇ Input
textaudioimagevideo
⬆ Output
textaudio

Benchmark results

1 benchmark
Speech Agent Arena
2. miejsce
📅 15 Sept 2026📄 Google DeepMind (blog)
Marketed as highly cost-effective; ranked second.

Sources and related pages