What is Gemini 3.1 Flash Live?
Gemini 3.1 Flash Live is the audio-to-audio model Google launched for its Live API. It streams speech, video frames and text over one WebSocket connection, replies in speech, lets users interrupt, calls your functions and can ground answers in Google Search. You pick a thinking level from minimal to high, and all of its audio is watermarked with SynthID.
Google now lists it as a legacy preview and recommends Gemini 3.8 Live, which costs the same and runs tools in the background. Google is also moving Search Live and Gemini Live to the 3.8 models, and no shutdown date is announced.
Is Gemini 3.1 Flash Live still available?
Yes, on the Gemini API and in Google AI Studio, but only as a legacy preview that Google wants you to replace with Gemini 3.8 Live.
- Status
- Legacy preview model, with no shutdown date announced.
- Replacement
- Gemini 3.8 Live, a stable release on the same price table, or Gemini 3.8 Live Extended Thinking for heavier reasoning.
- Google's own apps
- Google is moving Search Live to Gemini 3.8 Live and Gemini Live to Gemini 3.8 Live Extended Thinking.
- Moving over
- Change the model ID to gemini-3.8-live, drop the thinking level setting, and expect tools to run in the background by default.
Who is Gemini 3.1 Flash Live best for?
Gemini 3.1 Flash Live is best for teams already running voice agents on it through the Gemini Live API, especially phone agents that call tools, though new projects should start on Gemini 3.8 Live.
Apps already built on it
Nothing forces a move yet: Google has announced no shutdown date, and the upgrade to Gemini 3.8 Live costs the same when you are ready.
Phone and voice agents
Builders running call agents on it say pacing and interruptions feel closer to a person than stitched speech-to-text, LLM and text-to-speech pipelines.
Agents that call tools by voice
Google reports 90.8% on ComplexFuncBench Audio at high thinking, and answers can be grounded in Google Search.
Tuning thinking against latency
Its thinking level runs from minimal to high, a setting Gemini 3.8 Live no longer accepts.
What's included with Gemini 3.1 Flash Live?
Real-time speech and video input, spoken replies, interruptions, function calling, Google Search grounding, adjustable thinking, transcripts and tools for long sessions.
- Real-time audio and video
- Streams 16kHz audio, images at up to one frame a second and text over a WebSocket, and answers in 24kHz audio.
- Interruptions
- Users can cut in at any point and the model stops to listen.
- Function calling
- Calls your functions mid-conversation, one at a time, and waits for each result before it speaks again.
- Google Search grounding
- Answers can draw on live search results, with 5,000 free grounded requests a month on the paid tier.
- Thinking levels
- Minimal, low, medium or high, with minimal as the default for the lowest latency.
- Transcripts
- Text transcripts of both the user's speech and the model's spoken replies.
- Session tools
- Context window compression, session resumption and ephemeral tokens for long calls and browser clients.
How much does Gemini 3.1 Flash Live cost?
Gemini 3.1 Flash Live is free on the Gemini API's rate-limited free tier, and on the paid tier audio costs $3.00 per million input tokens and $12.00 per million output tokens, about $0.005 and $0.018 a minute.
- Free input and output tokens at lower rate limits
- Google Search grounding included
- Prompts and responses are used to improve Google's products
- Output price includes thinking tokens
- About $0.002 a minute
- About $0.005 a minute in and $0.018 a minute out
- 5,000 Google Search grounding requests a month free, then $14 per 1,000
Gemini 3.1 Flash Live shares this price table with Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, which Google recommends instead. There is no Batch or context caching option. The paid tier needs a linked billing account and at least $5 of prepaid credit, and paid prompts are not used to improve Google's products. Full pricing on ai.google.dev. We checked these prices on .
How do readers rate Gemini 3.1 Flash Live?
No one has rated Gemini 3.1 Flash Live yet. Be the first.
Sign in to rate. One rating per account.
Gemini 3.1 Flash Live pros and cons
It sounds natural and costs little for live voice, but it is a legacy preview with reported audio glitches, turn-detection bugs and blocking tool calls.
Pros
- Developers running phone agents on it say calls feel noticeably more natural than stitched speech-to-text, LLM and text-to-speech pipelines, with smoother pacing and interruptions.
- One connection replaces three services, and audio costs $3 / $12 per million tokens, a fraction of GPT-Realtime-2's $32 / $64.
- Strong spoken tool use on Google's tests: 90.8% on ComplexFuncBench Audio, against 71.5% for Gemini 2.5 Flash Native Audio.
- A free tier on the Gemini API and in Google AI Studio lets you prototype at no cost.
Cons
- Google calls it a legacy preview and points new work to Gemini 3.8 Live, which costs the same and is a stable release.
- Developers on Google's forum, GitHub and Reddit report audio that cuts out or lags, dropped calls, and turn detection that ignores the silence setting, with some reports open for months.
- Function calls block the conversation, and several builders say it sometimes fails to call a tool at all or stalls mid-call.
- Audio-only sessions stop at 15 minutes and audio-video sessions at 2 minutes unless you turn on context compression, and there is no context caching or Batch API.
What are the best Gemini 3.1 Flash Live alternatives?
Gemini 3.8 Live for new builds, GPT-Realtime-2 for OpenAI's stack with caching, and Grok Voice Think Fast 2.0 for a flat per-minute price.
Gemini 3.8 LiveFreeChoose Gemini 3.8 Live for any new build: Google's recommended replacement costs the same, is a stable release, runs tools in the background and switches between 97 languages mid-conversation.
GPT-Realtime-2PaidChoose it when you want OpenAI's Realtime API with prompt caching and adjustable reasoning, and can pay $32 / $64 per million audio tokens, more than ten times Gemini's input price.
Grok Voice Think Fast 2.0PaidChoose it when you want one flat rate to budget against; xAI charges $0.08 per minute of speech-to-speech.
Questions about Gemini 3.1 Flash Live
What is Gemini 3.1 Flash Live?
Gemini 3.1 Flash Live is Google's legacy preview model for real-time voice agents on the Gemini Live API: it hears speech, video and text, answers aloud, and calls tools mid-conversation.
Who is Gemini 3.1 Flash Live for?
Gemini 3.1 Flash Live is best for teams already running voice agents on it through the Gemini Live API, especially phone agents that call tools, though new projects should start on Gemini 3.8 Live.
Is Gemini 3.1 Flash Live free?
Yes, within limits. The Gemini API's free tier charges nothing for Gemini 3.1 Flash Live input and output, at lower rate limits and with your prompts used to improve Google's products. Paid audio costs $3.00 per million input tokens and $12.00 per million output tokens.
How much does Gemini 3.1 Flash Live cost?
Gemini 3.1 Flash Live's paid plans are Text at $0.75 / $4.50 per million input / output tokens, Image and video input at $1.00 per million input tokens and Audio at $3.00 / $12.00 per million input / output tokens.
What are the best Gemini 3.1 Flash Live alternatives?
The closest alternatives on Recaply are Gemini 3.8 Live, GPT-Realtime-2 and Grok Voice Think Fast 2.0.
Where can I try Gemini 3.1 Flash Live?
On its official website, blog.google.
Should I use Gemini 3.1 Flash Live or Gemini 3.8 Live?
Use Gemini 3.8 Live for new work. Google calls 3.1 Flash Live a legacy preview, both share one price table, and 3.8 Live adds background function calls and interleaved reasoning. Stay on 3.1 Flash Live only if you need its minimal-to-high thinking setting: 3.8 Live takes none, and 3.8 Live Extended Thinking starts at low.
How long can a Gemini 3.1 Flash Live session last?
Without context window compression, audio-only sessions end after 15 minutes and audio-video sessions after 2 minutes, and a single connection lasts about 10 minutes. Turn on compression, plus session resumption with tokens that stay valid for 2 hours, to keep longer calls going.
Does Google use my Gemini 3.1 Flash Live conversations to improve its products?
On the free tier, yes: Google's pricing page says free-tier content is used to improve its products. On the paid tier, which needs a linked billing account and at least $5 of prepaid credit, your prompts and responses are not used that way.
Can I use Gemini 3.1 Flash Live for plain transcription or text-to-speech?
It is built for live two-way conversation, not file jobs, and has no Batch API. Google offers separate models for those tasks: Gemini 3.5 Transcribe for speech-to-text with speaker labels and timestamps, and Gemini 3.8 Flash TTS for generating speech from text.
Our verdict
Gemini 3.1 Flash Live still works and is cheap, but new voice agents should start on Gemini 3.8 Live.
Google calls it a legacy preview, 3.8 Live costs exactly the same and runs tools in the background, and developers report audio glitches and turn-detection bugs on 3.1 that have stayed open for months.
- Skip it for
- Any new build, and agents that need tools to run while the conversation keeps going.
- Check first
- If you depend on its thinking level setting, test 3.8 Live on your own calls first, since it drops that control.
Gemini 3.1 Flash TTS
Gemini 3.8 Flash TTS
Gemini 3.8 Flash-Lite TTS
Gemini 3.5 Transcribe
MAI-Transcribe-2-Streaming
Scribe v2 Realtime