Voxtral Mini 3B (MLX, bfloat16)
Full-precision MLX-compatible weights for Mistral's Voxtral Mini speech-to-text model, optimized for Apple Silicon inference.
Voxtral Mini is built on Ministral 3B with state-of-the-art audio understanding capabilities. It supports transcription, translation, Q&A, summarization, and function calling directly from audio input.
Features
- 8 languages: English, Spanish, French, Portuguese, Hindi, German, Dutch, Italian
- Long-form audio: up to 30 min transcription, 40 min understanding
- 32k token context
- Voice-triggered function calling
Specifications
| Property | Value |
|---|---|
| Total Parameters | 4.68B (3B text decoder + 1.7B audio encoder) |
| Precision | bfloat16 |
| Download Size | 17.4 GB |
| License | Apache 2.0 |
Usage
Used by Mac X for on-device speech transcription on Apple Silicon via MLX.
See also: 8-bit quantized (5 GB) | 4-bit mixed quantized (3.2 GB)
License
Apache 2.0 — original model by Mistral AI.
- Downloads last month
- 1
Model size
5B params
Tensor type
F32
·
BF16 ·
Hardware compatibility
Log In to add your hardware
Quantized
Model tree for Aayush9029/Voxtral-Mini-3B-2507
Base model
mistralai/Voxtral-Mini-3B-2507