Runs in the browser · Speech to text
Run Voxtral Mini 4B Realtime in your browser
Voxtral Mini 4B Realtime downloads 2.64 GB over WebGPU (q4f16), or 3.47 GB over WebAssembly (q8), to transcribe speech to text entirely in the tab with Transformers.js. No install, no server.
Speech to text · onnx-community/Voxtral-Mini-4B-Realtime-2602-ONNX
- Parameters
- 4.43B
- Pipeline task
- automatic-speech-recognition
Reading
Built for realtime streaming transcription; the standard automatic-speech-recognition pipeline() call works for one-shot audio, but streaming needs custom chunking beyond the default pipeline. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today.
The largest speech to text download in the catalog (2.64 GB).
0.61 MB/M params against a 1.91 MB/M speech to text median: light for its parameter count.
Will it run in your browser?
Checking for WebGPU support…
All measured variants
| Variant | Size |
|---|---|
| q4f16 WebGPU pick | 2.64 GB |
| q4 | 2.73 GB |
| q8 WASM pick | 3.47 GB |
| fp16 | 4.54 GB |
| fp32 | 5.36 GB |
Sizes measured from the HuggingFace API file tree for onnx-community/Voxtral-Mini-4B-Realtime-2602-ONNX, not estimated.
Use it with Transformers.js
import { pipeline } from "@huggingface/transformers";
const pipe = await pipeline("automatic-speech-recognition", "onnx-community/Voxtral-Mini-4B-Realtime-2602-ONNX", {
device: "webgpu", // usually falls back to "wasm"; wrap in try/catch for production
dtype: "q4f16", // use "q8" for the WASM build
});
const result = await pipe(/* your input */);
Requires npm install @huggingface/transformers (or the CDN build).
Sources
Same job, different size
FAQ
WebGPU or WASM for Voxtral Mini 4B Realtime, in practice?
Chrome, Edge, and Safari 26+ run the WebGPU build (q4f16) on the GPU. Firefox has WebGPU on Windows and Apple Silicon Macs but not everywhere yet. Any browser without a WebGPU adapter falls back automatically to the WebAssembly build (q8) on CPU: same model, slower to load, slower to run.
What does q4f16 mean for Voxtral Mini 4B Realtime?
The WebGPU build here uses q4f16: 4-bit weights with some layers kept at 16-bit for stability: WebGPU's usual smallest clean build. The WASM fallback uses q8: 8-bit integers: about a quarter the size of fp32, with a smaller quality trade than 4-bit.
Sizes measured 2026-08-01 from the HuggingFace API. Last validated 2026-08-03. See all browser models or the methodology.