Gemini 3.8 Live Explained: Google's Real-Time Voice AI
In short: Gemini 3.8 Live is Google's AI model for real-time voice conversations - you talk, it listens and talks back with very little delay. It comes in two versions: Gemini 3.8 Live, and Gemini 3.8 Live Extended Thinking for conversations that need more reasoning. Both were released in September 2026 and are available in the Gemini API, Google Workspace and the Gemini app.
What is Gemini 3.8 Live?
Gemini 3.8 Live is an "audio-to-audio" model: it takes spoken input and replies with speech, in real time. Most AI models work with text: you type, they write back. A live voice model skips the typing. It listens to your voice, can take in images and video as you talk, and answers out loud - and you can interrupt it mid-sentence, the way you would a person.
Google describes the pair, in its announcement, as models that make voice interactions "more natural, fluid, and intelligent". Some of what it lists:
- 97 languages - it detects and switches between them mid-conversation.
- Visual input in near real time - it takes in images and video during the conversation for added context.
- Background tasks - it can run tools and API calls in the background and keep talking while they finish, instead of going silent.
What is the difference between Gemini 3.8 Live and Live Extended Thinking?
Live is built for fast, fluid conversation; Live Extended Thinking adds more background reasoning for harder requests, while keeping the conversation going. Google's announcement describes 3.8 Live as offering fast conversations with real-time visual and language support, and Extended Thinking as the version for complex tasks that need deeper reasoning. Both share the same prices and limits in the Gemini API.
Where can you use Gemini 3.8 Live?
According to Google's announcement, it is available:
- In the Gemini API - for developers building voice agents, through Google's Live API (model IDs
gemini-3.8-liveandgemini-3.8-live-extended-thinking). - In Google Workspace.
- In the Gemini app - for live voice conversations.
Google also offers Gemini 3.8 Live with Live Avatar in Gemini Enterprise, which adds an animated face that speaks in sync with the voice.
How much does Gemini 3.8 Live cost?
On the Gemini API's free tier it is free of charge; on the paid tier, text costs $0.75 per million input tokens and $4.50 per million output tokens, and audio costs $3 and $12 - or about half a cent per minute of audio in and under two cents per minute out. These figures are from Google's pricing page.
| What you send or receive | Paid tier price |
|---|---|
| Text input | $0.75 per 1M tokens |
| Audio input | $3.00 per 1M tokens, or $0.005 per minute |
| Image or video input | $1.00 per 1M tokens, or $0.002 per minute |
| Text output (including thinking) | $4.50 per 1M tokens |
| Audio output | $12.00 per 1M tokens, or $0.018 per minute |
Worked example. A 10-minute voice conversation with about 5 minutes of you talking and 5 minutes of the model talking costs roughly 5 × $0.005 + 5 × $0.018 = $0.115 on the paid tier.
The free tier comes with usage limits, and Google's pricing page notes that free-tier data is used to improve its products; paid-tier data is not.
What are Gemini 3.8 Live's specs?
From Google's model page:
- Inputs: text, images, audio and video. Outputs: text and audio.
- Input limit: 131,072 tokens. Output limit: 65,536 tokens.
- Function calling: supported, so it can use tools during a conversation.
- Thinking: supported, with reasoning interleaved into the conversation.
- Not supported: caching, code execution and structured outputs; the Batch API is not available for it.
It replaces Gemini 3.1 Flash Live, and Google's documentation includes a migration guide from that model.
How does a live voice model work?
A live model keeps an open connection and streams sound both ways, instead of sending one message and waiting for a full reply. In Google's Live API the app opens a WebSocket - a connection that stays open - and streams small chunks of your audio as you speak. The model starts answering as soon as it has enough, and streams its voice back. Because the connection stays open, it can notice when you interrupt and stop talking.
For how the model underneath turns sound and text into a reply, see how LLMs work.
Related
Frequently asked questions
What is Gemini 3.8 Live?+
Gemini 3.8 Live is Google's real-time voice AI model. It listens to speech, can take in images and video, and replies with speech, with low delay and support for being interrupted. It is available in the Gemini API, Google Workspace and the Gemini app.
What is the difference between Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking?+
Gemini 3.8 Live is tuned for fast, fluid conversation. Gemini 3.8 Live Extended Thinking adds more background reasoning for complex tasks while the conversation continues. Both have the same API prices and limits.
Is Gemini 3.8 Live free?+
On the Gemini API free tier it is free of charge, with usage limits. On the paid tier, audio costs $3 per million input tokens and $12 per million output tokens, about $0.005 and $0.018 per minute.
How many languages does Gemini 3.8 Live support?+
Google says it supports 97 languages and detects and switches between them automatically during a conversation.
When was Gemini 3.8 Live released?+
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking in September 2026, in its announcement of both models and in its September 2026 AI updates.
Can Gemini 3.8 Live use tools?+
Yes. Function calling is supported, and Google says it can run tools and API calls in the background and keep talking while they finish.
Related Guides
Enjoyed this? Get more in your inbox.
Up to ten of the week's AI news stories, by email once a week.