Voxtral Mini 3B 2507 — MLX 5-bit affine

MLX weights for Voxtral Mini 3B 2507, ready for local speech-to-text with speech-swift.

The FP16, INT5, and INT8 family passed the isolated WER, RTF, throughput, RSS, and physical-footprint compatibility gate on 2026-07-22 using English FLEURS.

Parameters	Precision	Sample rate	Languages
3B	5-bit affine	16 kHz	8

Files

File	Description
`model*.safetensors`	MLX model weights
`config.json` and `params.json`	Model and runtime configuration
`tekken.json`	Tekken tokenizer data

Usage

speech transcribe recording.wav --engine voxtral --model int5

import VoxtralASR

let model = try await VoxtralModel.load(
    "aufklarer/Voxtral-Mini-3B-2507-MLX-5bit"
)
let text = model.transcribe(
    audio: samples,
    sampleRate: sourceSampleRate,
    language: "en"
)

Benchmark

English FLEURS read speech, 80 utterances (759.56 seconds), Apple M5 Pro:

Variant	Bundle	WER	Mean RTF	Peak RSS
FP16	8.71 GiB	4.633%	0.1305	9,091 MiB
INT5	3.77 GiB	4.744%	0.0739	4,035 MiB
INT8	5.18 GiB	4.578%	0.0906	5,487 MiB

These read-speech measurements are not Artificial Analysis Conversational or ElevenLabs Scribe results.

Source and license

The base model is mistralai/Voxtral-Mini-3B-2507 and is released under Apache 2.0.

Model tree for aufklarer/Voxtral-Mini-3B-2507-MLX-5bit

Base model

mistralai/Voxtral-Mini-3B-2507

Quantized

(29)

this model

Collection including aufklarer/Voxtral-Mini-3B-2507-MLX-5bit

MLX Speech Models

Collection

Speech AI models for Apple Silicon via MLX. ASR, TTS, VAD, diarization, speaker embedding. • 92 items • Updated about 6 hours ago • 5

aufklarer
/

Voxtral-Mini-3B-2507-MLX-5bit