We are excited to introduce Gemini 3.1 Flash Live, now available to all developers through the Google Live API in Google AI Studio.
This update improves response latency, increases uptime, and enhances the realism of synthesized speech setting a technical benchmark for voice-first AI applications.
Enjoy Improved Speed, Reliability, and Overall Quality
In real-time conversations, even small delays can disrupt the natural flow users expect. The new model captures tone, emphasis, and implication more accurately, leading to several key improvements.
- Higher task completion in noisy environments. Their model is now much better at using external tools and sharing information during live conversations. It can distinguish between important speech and background sounds like traffic or TV, so it stays reliable and follows instructions even in busy settings.
- Instruction-following is improved via advanced intent parsing and context retention. The model maintains compliance with agent guidelines, even as conversational topics shift and tasks require adaptive understanding.
- Dialogue latency is minimized through model optimization, allowing rapid detection of nuanced audio features such as pitch and pace. This results in conversations that emulate natural human timing and turn-taking more effectively than prior versions.
- The model’s multilingual capabilities now support real-time voice interactions in over 90 languages, enabled by parallelized training and language-specific acoustic modeling.
Watch the Gemini Live API at Work
Developers are already using Gemini Flash Live models to build voice agents that communicate naturally and act reliably. Here are some real-world apps powered by the model.
Take Advantage Of A Growing Range Of Integrations
The Live API is designed for production use and can handle a variety of inputs, including live video, streams, and on-demand phone calls.
If your system needs web, IC scaling, or global edge routing, check out our partner integrations to make it easier to build real-time voice and video agents.
Start Building With The Live API
Gemini 3.1 Flash Live is now available through the Gemini App API, and Google AI Studio developers can use the Gemini Live API to add the model to their apps. See developer documentation to learn how you can build real-time agents.
- Find details on multilingual support, tool use, session management, and more in the Gemini Live API documentation.
- Discover sample voice experiences you can build today with Gemini Drive API examples.
- Enhance your coding agents with the Gemini Live API SQL.
Start using the Google Gen AI SDK.
import asyncio from google import genai
client = genai.Client(api_key=”YOUR_API_KEY”)
model = “gemini-3.1-flash-live-preview” config = {“response_modalities”: [“AUDIO”]}
async def main(): async with client.aio.live.connect(model=model, config=config) as session: print(“Session started”) # Send content…
if name == “main”: asyncio.run(main())
Source: Build real-time conversational agents with Gemini 3.1 Flash Live
