← All models

LFM2-Audio-1.5B

Liquid AI · 2025-10-01 · 1.6B parameters

LFM2-Audio-1.5B is Liquid AI's October 1, 2025 end-to-end audio foundation model for low-latency speech conversation: the LFM2-1.2B backbone with a 115M FastConformer audio encoder (from canary-180m-flash) and an RQ-transformer that generates Mimi audio tokens, taking text or audio and answering in text, speech or both. Its main weights hold 1,470,308,488 BF16 parameters, and the repository also ships Kyutai's Mimi codec (96,151,393 parameters, CC-BY-4.0), for 1,566,459,881 in the released checkpoint. The card gives a 32,768-token context and English support.

Benchmark scores