Model Specifications
Voxtral Model Family
Choose the right model for your use case. All models feature automatic language detection, built-in Q&A capabilities, and Apache 2.0 licensing.
Recommended
Voxtral Largev1.0
Our flagship model with 32K token context window and state-of-the-art accuracy
95.2%
Accuracy
32K tokens
Context
15+
Languages
7B parameters
Parameters
Key Features
- 32K token context window (40+ minutes)
- Built-in Q&A and summarization
- Function calling from voice
- Automatic language detection
- Real-time processing support
Benchmark Results
Common Voice95.2%
vs 91.7% Whisper large-v3
FLEURS93.8%
vs 89.4% Whisper large-v3
LibriSpeech96.1%
vs 94.2% Whisper large-v3
Voxtral Basev1.0
Balanced performance and efficiency for most use cases
92.8%
Accuracy
16K tokens
Context
15+
Languages
1.5B parameters
Parameters
Key Features
- 16K token context window (20+ minutes)
- Built-in Q&A and summarization
- Automatic language detection
- Optimized for edge deployment
Benchmark Results
Common Voice92.8%
vs 89.1% Whisper base
FLEURS90.5%
vs 86.2% Whisper base
LibriSpeech93.4%
vs 91.8% Whisper base
Voxtral Smallv1.0
Lightweight model optimized for mobile and edge devices
89.4%
Accuracy
8K tokens
Context
10+
Languages
244M parameters
Parameters
Key Features
- 8K token context window (10+ minutes)
- Automatic language detection
- Mobile and edge optimized
- Low memory footprint
Benchmark Results
Common Voice89.4%
vs 85.7% Whisper small
FLEURS86.2%
vs 82.1% Whisper small
LibriSpeech90.8%
vs 88.3% Whisper small
Supported Languages
All models support automatic language detection across these languages
English
Spanish
French
Portuguese
Hindi
German
Dutch
Italian
Chinese
Japanese
Korean
Russian
Arabic
Turkish
Polish
Quick Start
Get started with Voxtral models using our API or self-hosted deployment
API Usage
curl -X POST https://api.mistral.ai/v1/audio/transcriptions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: multipart/form-data" \
-F "file=@your_audio.wav" \
-F "model=voxtral-large" \
-F "response_format=verbose_json"Self-hosted Deployment
docker run -p 8000:8000 \
-v ./models:/models \
voxtral/voxtral-server:latest \
--model voxtral-largeReady to Try Voxtral?
Experience the power of our speech AI models with our interactive demo or start building with our API.