Breaking

Google Launches Gemini 3.8 Live and Extended Thinking for Real-Time Voice Agents

September 16, 2026 at 04:32 EDT

Foundation Models
AI Agents

Google has released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two audio-to-audio models aimed at making voice agents faster, more capable and less prone to awkward pauses during complex tasks. The launch puts Google's latest live dialogue stack into the Gemini API, Google AI Studio, Search Live, Gemini Live and selected Workspace experiences as the market for real-time conversational AI becomes more competitive.

Sep. 15, 2026 · Google

Gemini voice agents get background reasoning

Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking bring near real-time voice, vision context, multilingual switching and background tool use to Google's live AI stack.

supported languages

82.6

Speech-to-Speech score

68.6%

tau-Voice result

$0.018

audio output per minute

Agentic benchmark split

68.6%

tau-Voice

35.1%

tau-Voice-banking

The two percentages are shown on the same scale: one stacked block represents roughly 10 percentage points.

Listen and see

Audio, visual context and live inputs shape the response.

→

Call tools

API requests can run in the background.

Keep talking

The conversation continues while work completes.

What developers gain

Better fit for tutoring, support, live troubleshooting, booking flows and speech-driven coding.

What changes in apps

Extended Thinking needs non-blocking tools and state tracking with interaction_status.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.