Open SourceMedium importance
Liquid AI releases the weights of LiquidCodec, a 74-million-parameter speech codec that encodes 361 times faster than real time on a single CPU thread
Liquid AI published on Hugging Face LiquidCodec, the codec behind the LFM3-Audio-2B and LFM3-Omni-2B models. With about 74 million parameters, it turns 24 kHz speech into tokens a language model can predict and converts them back into audio, with 12 token streams every 80 milliseconds. It replaces Kyutai's Mimi. In the company's own tests, on one CPU thread of a MacBook Pro M4 Max, it encodes about 361 times faster than real time and decodes about 87 times faster, and it had the best reconstruction score on the eight test sets against X-Codec 2, NanoCodec and Mimi, although NanoCodec wins on PESQ in seven of them. There are GGUF versions for llama.cpp.
Published in edition no. 27 of Noticiário do Joãozinho ·
Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese
Original source
https://www.marktechpost.com/2026/11/12/liquid-ai-liquidcodec-streaming-audio-codec/