KRAFTON Releases A.X-K2 Speech Model Supporting Any-to-Any Modality

KRAFTON · hf · 2026-08-02

KRAFTON's A.X-K2-Raon-Speech-21B-A3B model has landed on the Hugging Face trending list.

The model features 21B total parameters with 3B active parameters, utilizing a Mixture-of-Experts (MoE) architecture. It is designed as an any-to-any pipeline, with core tags covering speech, audio, automatic speech recognition (ASR), and text-to-speech (TTS).

Original post →

More from Models

Models channel →