Conversational voice agents
Convert live caller speech into text quickly enough to feed an agent response loop.
Independent tool overview
Scribe v2 Realtime at a glance
Scribe v2 Realtime is ElevenLabs' streaming speech-to-text model for voice agents, live captions, meeting assistants, and other latency-sensitive applications. It returns partial transcripts in roughly 150 milliseconds, supports more than 90 languages, and is available through the ElevenLabs API and ElevenAgents.
Visit the official Scribe v2 Realtime site ↗
- Best for
- Voice agents, live captions, and streaming transcription
- Latency
- About 150ms for partial transcripts
- Languages
- 90+ with automatic detection
- Standard API rate
- $0.39 per audio hour
- Access
- API and ElevenAgents
Overview
What Scribe v2 Realtime is
Scribe v2 Realtime is a developer-facing transcription model rather than a standalone meeting-notes app. Applications stream microphone or telephony audio over a WebSocket and receive partial text while a person is speaking, followed by committed transcript segments when speech is finalized.
The model is designed for conversational responsiveness. It combines automatic language detection, voice activity detection, manual commit controls, previous-text conditioning, precise word timestamps, and entity detection. ElevenLabs says it can handle language switching within a conversation and supports PCM audio from 8kHz to 48kHz plus mu-law encoding.
Choose it when low latency matters more than the fuller post-processing available from a batch model. Teams still need to build the surrounding capture, consent, storage, retry, and transcript-review experience themselves.
Use cases
Who Scribe v2 Realtime is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Realtime captions
Generate partial and finalized captions for meetings, broadcasts, classes, or accessibility interfaces.
Meeting and call products
Add live transcripts to a custom application while retaining control over the user interface and downstream workflow.
Multilingual applications
Support conversations across more than 90 languages, including language changes within a session.
Capabilities
Core Scribe v2 Realtime features
Low-latency streaming
Returns partial transcript updates in roughly 150 milliseconds through a WebSocket connection.
90+ languages
Automatically recognizes supported languages and can follow multilingual speech within the same conversation.
Partial and committed transcripts
Applications can show provisional words immediately and replace them with stable transcript segments after a commit.
Voice activity detection
Silence detection can segment speech automatically, while manual commit gives developers explicit finalization control.
Word-level timestamps
Precise timing metadata supports synchronized captions and downstream audio navigation.
Text conditioning
Previous transcript context can be supplied after a reconnect so the model continues with better linguistic continuity.
Entity detection
The current model reference lists detection across 65 entity types for applications that need structured information from live speech.
Process
How the Scribe v2 Realtime workflow works
Step 1
Create a secure session
A server can create a single-use token for client-side streaming so the permanent ElevenLabs API key is not exposed in the browser.
Step 2
Stream supported audio
Send microphone or telephony audio chunks to the realtime WebSocket in a supported PCM or mu-law format.
Step 3
Render partial text
Display partial transcript events for immediacy, understanding that predictions can still change.
Step 4
Commit and process
Use automatic voice activity detection or manual commit, then store or process only the finalized transcript according to the application's privacy policy.
Cost
Scribe v2 Realtime pricing and free plan
ElevenLabs lists Scribe v2 Realtime at a standard API rate of $0.39 per audio hour. Subscription plans range from Free to Business and include different amounts of usage; the product page also advertises rates of $0.28 per hour or lower on annual Business plans. Taxes and any surrounding agent, storage, or application infrastructure are separate.
Standard API usage
$0.39/audio hour
The published Scribe v2 Realtime model rate for free and self-serve API accounts.
- Metered by audio duration
- Available through the API
- Taxes excluded
Platform subscriptions
$0-$990/month
Free, Starter, Creator, Pro, Scale, and Business plans include different realtime transcription allowances alongside other ElevenLabs products.
- Free plan available
- Included hours depend on plan
- Overages and other API products are priced separately
Annual Business volume
$0.28/audio hour or lower
A lower rate advertised for annual Business volume; actual effective cost depends on the selected commitment and usage.
- Annual commitment
- Designed for larger workloads
Enterprise
Custom
For negotiated limits, support, data controls, and compliance requirements.
- Custom concurrency and terms
- Zero-retention options
- HIPAA BAA available through Sales
Pricing checked . Check current pricing at the source ↗
Assessment
Scribe v2 Realtime strengths and limitations
Where it stands out
- Fast partial transcripts are suitable for responsive voice experiences
- Broad multilingual support reduces the need to route each language to a separate model
- WebSocket events separate provisional text from finalized transcript segments
- Supports both microphone-quality PCM and mu-law telephony audio
- Can be used inside ElevenAgents or as a standalone transcription API
What to consider
- It is an API model, so teams must build recording controls, consent, transcript storage, review, and end-user interfaces.
- Partial transcripts are predictive and can change before the segment is committed.
- The product FAQ says realtime speaker diarization is not currently a priority and dual-channel support is not planned.
- Zero-retention mode is limited to eligible enterprise or trial accounts rather than every API session.
- HIPAA workloads require a signed Business Associate Agreement with ElevenLabs before deployment.
- Costs can exceed the transcription rate once voice-agent inference, synthesis, telephony, storage, and observability are added.
Compare
Scribe v2 Realtime alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Consumer
Cohere Transcribe
An open-source speech-recognition option for teams prioritizing model control and self-hosting.
Explore Cohere Transcribe →Content Creator
Gemini 3.5 Transcribe
A Google transcription option for teams already building around the Gemini API ecosystem.
Explore Gemini 3.5 Transcribe →Miscellaneous
Voxtral Transcribe 2
A Mistral speech-to-text family that includes an open-weights realtime model for greater deployment flexibility.
Explore Voxtral Transcribe 2 →Questions
Scribe v2 Realtime FAQs
What is Scribe v2 Realtime?
It is ElevenLabs' streaming speech-to-text model for applications such as voice agents, live captions, and meeting assistants. It sends partial and committed transcripts over a realtime API connection.
How fast is Scribe v2 Realtime?
ElevenLabs documents roughly 150 milliseconds of latency for partial transcription. End-to-end application latency will also include network, audio buffering, agent inference, and any speech-synthesis time.
How much does Scribe v2 Realtime cost?
The standard published API rate is $0.39 per audio hour as of August 29, 2026. Annual Business pricing can reach $0.28 per hour or lower, while subscription plans include different usage allowances.
How many languages does it support?
ElevenLabs lists support for more than 90 languages with automatic language recognition and multilingual switching.
Does it identify different speakers?
The realtime product FAQ says speaker diarization is not currently a priority. If speaker labels are essential, verify current support or use a batch transcription workflow designed for diarization.
Can it run directly in a browser?
Yes, client-side microphone streaming is supported, but the browser should use a short-lived single-use token created by a trusted server rather than exposing the permanent API key.
Bottom line
Our Scribe v2 Realtime verdict
Scribe v2 Realtime is a compelling fit for developers who need responsive multilingual transcription and already plan to build the surrounding application. Test it with real accents, noise, telephony audio, and domain vocabulary before committing, and use batch Scribe v2 instead when finalized transcript richness matters more than immediacy.
Visit Scribe v2 Realtime website ↗Similar Consumer Tools
View all Consumer AI tools →
ChatGPT Atlas
Atlas - OpenAI's new web browser built with ChatGPT at its core
Grok 4.1
Grok 4.1 - xAI's latest update to Grok with improved creative abilities, emotional intelligence, and real-world usability.
Apps SDK
Apps SDK - Chat with and build apps directly in ChatGPT
Gemini 3
Gemini 3 - Google's most intelligent model release to date, taking the top spot industry-wide across major benchmarks for multimodal understanding, agentic abilities, reasoning, and more.
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.