Meeting and interview archives
Create searchable transcripts with speaker labels and word-level timestamps from long recordings.
Independent tool overview
Voxtral Transcribe 2 at a glance
Voxtral Transcribe 2 is Mistral's speech-to-text family, pairing a low-cost batch API with a 4B real-time model that is also available as Apache-licensed open weights for private or edge deployment.
Visit the official Voxtral Transcribe 2 site ↗
- Product type
- Batch and real-time speech-to-text
- Languages
- 13
- Batch API
- $0.003 per audio minute
- Realtime API
- $0.006 per audio minute
- Open model
- Voxtral Realtime 4B
- Open-model license
- Apache 2.0
Overview
What Voxtral Transcribe 2 is
Voxtral Transcribe 2 includes two distinct products. Voxtral Mini Transcribe V2 handles recorded audio with speaker labels, word timestamps and vocabulary hints, while Voxtral Realtime transcribes a live audio stream with configurable latency.
Both models support 13 languages: English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian and Dutch. The batch model can process recordings up to three hours in one request; the real-time model is aimed at live captions, voice agents and call-assist systems.
The deployment choice matters. The batch model is a hosted Mistral API at $0.003 per audio minute. Realtime costs $0.006 per minute through the API or can be self-hosted from the 4B Apache 2.0 checkpoint, shifting cost and privacy responsibility to the operator.
Use cases
Who Voxtral Transcribe 2 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Voice agents and live assistants
Stream text with adjustable delay so downstream models can respond while a person is still speaking.
Privacy-sensitive engineering teams
Self-host the Realtime checkpoint when audio must remain inside controlled infrastructure.
Capabilities
Core Voxtral Transcribe 2 features
Batch transcription
Processes recorded audio through Voxtral Mini Transcribe V2, including files up to three hours per request.
Streaming transcription
Voxtral Realtime uses a native streaming architecture with delay configurable for latency-versus-accuracy trade-offs.
Speaker diarization
Labels speakers and provides start and end times for multi-party recordings in the batch workflow.
Word-level timestamps
Returns timing for individual words to support subtitles, search, playback alignment and review tools.
Context biasing
Accepts up to 100 names, technical terms or phrases to improve expected spellings; English support is the most mature.
Noise robustness
Is designed for difficult environments such as call centers, factory floors and field recordings.
Open real-time deployment
Makes the 4B Realtime weights available under Apache 2.0 for edge, private-cloud or on-premises use.
Process
How the Voxtral Transcribe 2 workflow works
Step 1
Choose batch or streaming
Use batch for completed recordings and richer metadata; use Realtime when partial text must arrive during the conversation.
Step 2
Prepare audio and context
Normalize the input where appropriate and provide a compact list of important names or domain terms.
Step 3
Transcribe and retain confidence cues
Store timestamps, speaker labels and source-audio links so users can inspect uncertain passages.
Step 4
Review consequential content
Verify names, numbers, commitments and regulated information against the audio before downstream action or publication.
Cost
Voxtral Transcribe 2 pricing and free plan
Mistral bills the hosted models by audio minute. The open Realtime checkpoint has no model license fee, but self-hosting adds compute and operations costs.
Voxtral Mini Transcribe V2 API
$0.003/minute
Hosted batch transcription with diarization, timestamps and context biasing.
- About $3 per 1,000 audio minutes
- Up to three hours per request
- 13 supported languages
- Model identifier: voxtral-mini-latest
Voxtral Realtime API
$0.006/minute
Hosted streaming transcription for live applications.
- About $6 per 1,000 audio minutes
- Configurable transcription delay
- 13 supported languages
- Model identifier: voxtral-mini-transcribe-realtime-2602
Voxtral Realtime open weights
$0 license fee
Self-host the 4B BF16 checkpoint under Apache 2.0.
- Infrastructure and engineering not included
- Operator owns scaling and monitoring
- Useful for private or edge deployment
- Review model and dependency licenses
Pricing checked . Check current pricing at the source ↗
Assessment
Voxtral Transcribe 2 strengths and limitations
Where it stands out
- Low published per-minute pricing for the hosted batch API
- One family covers both completed recordings and live streams
- Speaker labels, word timestamps and vocabulary hints support production workflows
- Thirteen languages are supported across both models
- Realtime latency can be tuned against accuracy needs
- Apache-licensed Realtime weights create a credible private-deployment path
What to consider
- Accuracy varies with language, accent, microphone quality, noise and domain vocabulary
- The batch model typically transcribes only one speaker when speech overlaps
- Context biasing is optimized for English; other-language support is experimental
- The lowest-latency setting can sacrifice accuracy compared with a longer delay
- Speaker labels and timestamps still need review for legal, medical or compliance records
- The open checkpoint covers Realtime, not the hosted batch model
- Private deployment requires GPU capacity, serving expertise, security controls and ongoing evaluation
- Recording consent and privacy obligations remain the customer's responsibility
Compare
Voxtral Transcribe 2 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Content Creator
AssemblyAI
A mature hosted speech-to-text API with a broader menu of audio-intelligence features.
Explore AssemblyAI →Marketing
ElevenLabs
A broader voice platform that includes speech recognition alongside synthesis and agent tooling.
Explore ElevenLabs →Content Creator
Transistor AI Transcription
A simpler creator workflow for transcribing and editing podcast episodes rather than building an API product.
Explore Transistor AI Transcription →Questions
Voxtral Transcribe 2 FAQs
What is the difference between Voxtral Mini Transcribe V2 and Voxtral Realtime?
Mini Transcribe V2 is the hosted batch model for completed recordings, with diarization, context biasing and word timestamps. Realtime is the low-latency streaming model and is also available as open weights.
How much does Voxtral Transcribe 2 cost?
The batch API costs $0.003 per audio minute and the Realtime API costs $0.006 per minute. That is about $3 or $6 respectively for 1,000 minutes, before other infrastructure or application costs.
Which languages does Voxtral support?
English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian and Dutch.
Is Voxtral Realtime open source?
Mistral releases the 4B Realtime weights under the Apache 2.0 license. The hosted batch model is accessed through Mistral's API rather than downloaded from that checkpoint.
How low is Voxtral Realtime latency?
Mistral advertises configurable delay down to below 200 milliseconds. Lower delay is not automatically better; choose a setting by measuring accuracy and responsiveness on your audio.
Can Voxtral be used for compliance transcripts?
It can supply timestamps and speaker labels, and Mistral offers private deployment options. Human verification, recording consent, retention policy and the rest of a compliant process are still required.
Bottom line
Our Voxtral Transcribe 2 verdict
Voxtral Transcribe 2 is a strong option for teams that need both economical batch transcription and a real-time path. The open 4B streaming model is the differentiator, while high-stakes transcripts still require source-audio review and a carefully governed deployment.
Visit Voxtral Transcribe 2 website ↗Similar Miscellaneous Tools
View all Miscellaneous AI tools →
SAM Audio
SAM Audio - Meta's model to separate any sound from an audio or audio-visual source using natural language prompts
Model Council
Model Council - Perplexity's new tool for querying and synthesizing outputs from multiple models into a single answer
GLM-4.6V
GLM-4.6V - Zhipu AI's open-source multimodal model family with native tool use capabilities
.png)
Raven-1
Raven-1 - Tavus's real-time emotional perception model for AI conversations
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.