Google / Gemini Live

Gemini 2.5 Flash Live Native Audio

Enterprise native-audio release for real-time voice conversations, transcription and function calling, with configurable affective and proactive responses.

Speech & voice Transcription

deprecatedClosed weightsVoice conversationSpeech-to-speechTranscription

Official documentation (opens in a new tab)Reviewed 14 Oct 2026
30 of 33 specification fields collected

Input token limit128K (provider-rounded)See specification notes below

Maximum output64K (provider-rounded)See specification notes below

Input billingSee ratesGoogle Enterprise Agent Platform

Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Input token limit: 128K (provider-rounded)Source for Input token limit (opens in a new tab)
Maximum output: 64K (provider-rounded)Source for Maximum output (opens in a new tab)
Input modalities: Text, Image, Audio, VideoSource for Input modalities (opens in a new tab)
Output modalities: Text, AudioSource for Output modalities (opens in a new tab)
Reasoning: Not supportedSource for Reasoning (opens in a new tab)
Tool calling: SupportedSource for Tool calling (opens in a new tab)
Structured output: Not supportedSource for Structured output (opens in a new tab)
Streaming: SupportedSource for Streaming (opens in a new tab)
Batch processing: Not supportedSource for Batch processing (opens in a new tab)
Fine-tuning: Not supportedSource for Fine-tuning (opens in a new tab)
Context caching: Not supportedSource for Context caching (opens in a new tab)
Architecture: Not collected
Parameters: Not collected
Knowledge cutoff: Not collected
Languages: View all 24 entries
Arabic (Egyptian) (ar-EG)
Bengali (Bangladesh) (bn-BD)
Dutch (Netherlands) (nl-NL)
English (India) (en-IN & hi-IN bundle)
English (US) (en-US)
French (France) (fr-FR)
German (Germany) (de-DE)
Hindi (India) (hi-IN)
Indonesian (Indonesia) (id-ID)
Italian (Italy) (it-IT)
Japanese (Japan) (ja-JP)
Korean (Korea) (ko-KR)
Marathi (India) (mr-IN)
Polish (Poland) (pl-PL)
Portuguese (Brazil) (pt-BR)
Romanian (Romania) (ro-RO)
Russian (Russia) (ru-RU)
Spanish (US) (es-US)
Tamil (India) (ta-IN)
Telugu (India) (te-IN)
Thai (Thailand) (th-TH)
Turkish (Turkey) (tr-TR)
Ukrainian (Ukraine) (uk-UA)
Vietnamese (Vietnam) (vi-VN)
Source for Languages (opens in a new tab)
Release date: 2025-12-12Source for Release date (opens in a new tab)
Code execution: Not supportedSource for Code execution (opens in a new tab)
Computer use: Not supportedSource for Computer use (opens in a new tab)
URL context: Not supportedSource for URL context (opens in a new tab)
Search grounding: SupportedSource for Search grounding (opens in a new tab)
Live API: SupportedSource for Live API (opens in a new tab)
Concurrent requests: 1,000Source for Concurrent requests (opens in a new tab)
Session duration: 10-minute default conversation duration; extendableSource for Session duration (opens in a new tab)
Audio formats: Live stream: raw signed 16-bit little-endian PCM, mono. Broader upload MIME types listed in the card are not the WebSocket wire format.Source for Audio formats (opens in a new tab)
Audio sample rates: Input 16 kHz; output 24 kHzSource for Audio sample rates (opens in a new tab)
Image input limits: 3,000 images/prompt, 7 MB inline/console or 30 MB Cloud Storage. PNG/JPEG/WEBP/HEIC/HEIF.Source for Image input limits (opens in a new tab)
Source video limits: Live input uses 1 FPS JPEG frames; card lists standard 768 × 768 resolution. No live-avatar video output for this release.Source for Source video limits (opens in a new tab)
Voice controls: Affective dialog and proactive audio require explicit configuration on 2.5. Proactive audio is preview. Speech start sensitivity low, end sensitivity high, prefix padding zero by default.Source for Voice controls (opens in a new tab)
Stock voice catalog: 30 predefined voicesSource for Stock voice catalog (opens in a new tab)
Language switching within audio: Natural language switching during conversation; system instructions can constrain spoken languages.Source for Language switching within audio (opens in a new tab)
Documented tool behavior: Asynchronous and synchronous function calls. An interrupting tool response may overlap the user’s speech; newer 3.8 instead waits for the user to finish.Source for Documented tool behavior (opens in a new tab)
API compatibility: Card lists GenerateContent and Gemini Live API; excludes Interactions, CountTokens and Chat Completions.Source for API compatibility (opens in a new tab)
Published availability regions: View all 13 entries
us-central1
us-east1
us-east4
us-east5
us-south1
us-west1
us-west4
europe-central2
europe-north1
europe-southwest1
europe-west1
europe-west4
europe-west8
Source for Published availability regions (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Google Enterprise Agent Platform

Provider source (opens in a new tab)

API model ID: gemini-live-2.5-flash-native-audio
Access: Google Cloud project with billing and Agent Platform API enabled. Standard PayGo or Provisioned Throughput; regional, concurrency and IAM limits apply.
Regions: us-central1, us-east1, us-east4, us-east5, us-south1, us-west1, us-west4, europe-central2, europe-north1, europe-southwest1, europe-west1, europe-west4, europe-west8
provisionedThroughput: SupportedSource for provisionedThroughput (opens in a new tab)
standardPayGo: SupportedSource for standardPayGo (opens in a new tab)
Batch processing: Not supportedSource for Batch processing (opens in a new tab)
Context caching: Not supportedSource for Context caching (opens in a new tab)
liveAvatar: Not supportedSource for liveAvatar (opens in a new tab)
proactiveAudio: SupportedSource for proactiveAudio (opens in a new tab)
affectiveDialog: SupportedSource for affectiveDialog (opens in a new tab)
sessionResumption: Not collected

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
Category	Usage	Rate	Unit
Live API inference	Text input¹	$0.50	1M text input tokens
Live API inference	Image/video input¹	$3.00	1M image/video input tokens
Live API inference	Audio input¹	$3.00	1M audio input tokens
Live API inference	Text output¹	$2.00	1M text output tokens
Live API inference	Audio output¹	$12.00	1M audio output tokens

Rate conditions

Enterprise Gemini 2.5 Flash Live API standard rates. Modalities are billed separately; batch rates are not available. No text-only question quote is assigned to this live voice endpoint.

Pricing source (opens in a new tab)

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

Google Enterprise Agent Platform

gemini-live-2.5-flash-native-audio

Not supported

Support evidence (opens in a new tab)Caching reviewed 14 Oct 2026

Voice choices

30 voices

Choose a voice within Gemini 2.5 Flash Live Native Audio. These are API voice identifiers, not additional model releases. Source links document each voice.

Find a voice

Published voice catalog
Voice / API ID	Language & accent	Delivery
Zephyr`Zephyr`	Multilingual	BrightVoice source (opens in a new tab)
Kore`Kore`	Multilingual	FirmVoice source (opens in a new tab)
Orus`Orus`	Multilingual	FirmVoice source (opens in a new tab)
Autonoe`Autonoe`	Multilingual	BrightVoice source (opens in a new tab)
Umbriel`Umbriel`	Multilingual	Easy-goingVoice source (opens in a new tab)
Erinome`Erinome`	Multilingual	ClearVoice source (opens in a new tab)
Laomedeia`Laomedeia`	Multilingual	UpbeatVoice source (opens in a new tab)
Schedar`Schedar`	Multilingual	EvenVoice source (opens in a new tab)

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Google publishes this named proprietary hosted release and selector without downloadable weights. This is a version-level catalog record, not a claim of a new base architecture or distinct weights from another host or preview. No explicit selector-alias equivalence is established in the reviewed card.

Classification evidence (opens in a new tab)

Lifecycle dates

Enterprise release12 Dec 2025Source (opens in a new tab)

Scheduled Enterprise retirement13 Dec 2026Source (opens in a new tab)

Sources reviewed

google enterprise veo31 (opens in a new tab)Reviewed 14 Oct 2026

google enterprise embedding2 card (opens in a new tab)Reviewed 14 Oct 2026

google enterprise live25 (opens in a new tab)Reviewed 14 Oct 2026

google enterprise live migration (opens in a new tab)Reviewed 14 Oct 2026

google enterprise live voices (opens in a new tab)Reviewed 14 Oct 2026

google enterprise live38 guide (opens in a new tab)Reviewed 14 Oct 2026

google enterprise pricing (opens in a new tab)Reviewed 14 Oct 2026

google enterprise lifecycle (opens in a new tab)Reviewed 14 Oct 2026

Model notes

Coverage includes reviewed Gemini API and Enterprise releases, host-specific access/pricing/cache details, original Flash and Argon reports, and historical Gemini, PaLM, Codey, Gecko and Imagen versions. Earlier previews and retired selectors are kept distinct. Exact historical specifications and further releases remain incomplete; one host’s shutdown does not establish worldwide retirement.

Official Google provider icon used as fallback; no unique model artwork is claimed.

Host/version-specific API limits are recorded separately from the Developer API preview. The available sources do not prove whether their underlying weights are identical.

Scheduled retirement is December 13, 2026 on Enterprise. This is not a claim that every Gemini API native-audio preview retires on that date.

Audio input/output and transcripts have different token prices. A text-only question estimate does not represent a live voice session.

The voice guide also describes limited-access custom voices, but it does not establish support for this exact 2.5 model; voice cloning is left unverified.

Check against your workload

Try representative inputs, measure task success and total cost, then check the exact endpoint’s data policy.

Read our methodology