# Zyphra's ZONOS3-Mini generates speech in real time using partial activation

> Original title: ZONOS3-Mini activates only part of the model at each generation step. Our new audio codec also cuts the tokens needed to represent speech...

- Source: Zyphra
- Published: 2026-11-09T22:07:54.000Z
- Bad Influence: https://badinfluence.news/items/samnjlh8prknpe3nd7axbw9eb
- Original: https://x.com/ZyphraAI/status/2119914904786526418

## Summary

Zyphra says ZONOS3-Mini activates only part of the model at each generation step. Its new audio codec reduces the tokens needed to represent speech from 774 to 150 per second, which the company says enables real-time generation across a wide range of hardware.

## Post

ZONOS3-Mini activates only part of the model at each generation step. Our new audio codec also cuts the tokens needed to represent speech from 774 to 150 per second.

This lets us generate in real-time across a vast range of hardwares.