Ant releases Ming-3.0 Audio, an open-source audio model with speech-to-speech capabilities

·2026.10.21 14:12

Key point

The model accepts audio and text inputs to generate spoken or written responses, with Flash and tiny checkpoints available.

Ant released the Ming-3.0 Audio model family under the MIT license, including weights and code.
The model accepts audio combined with text instructions and produces either written or spoken responses.
Two Mixture of Experts (MoE) checkpoints are provided: Flash with 126B total and 6.5B active parameters, and tiny with 9.3B total and 2.7B active parameters.
Demonstrations show the model analyzing earnings calls, identifying sounds, and engaging in spoken conversations based on user input.

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.