inclusionAI releases Ming-3.0-audio-flash, a 126B-parameter native audio-language model

·2026.10.19 23:27

Key point

The model activates 6.5B parameters per inference step and supports speech recognition, audio reasoning, and synthesized speech responses under an MIT license.

Details

inclusionAI released Ming-3.0-audio-flash, a native audio-language Mixture-of-Experts model with 126B total parameters and 6.5B activated per inference step.
The model accepts audio and text inputs to perform speech recognition, spoken question answering, audio understanding, and audio reasoning.
It supports Chinese and English, including varied accents, dialects, noisy environments, and multi-speaker recordings.
Outputs include text or text with optional synthesized speech using built-in or reference voices.
The model is available under an MIT license and is hosted on Hugging Face, ModelScope, and GitHub.

open-source moe speech-recognition audio-ai inclusionai ming-3.0

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.