← All models

LFM3-Omni-2B

Liquid AI · 2026-10-27 · 2.2B parameters

LFM3-Omni-2B is Liquid AI's first omni-modal model, released with open weights on October 27, 2026 in the first LFM3 family. One end-to-end model takes text, images, video with or without its soundtrack, and 16 kHz audio, and answers in text, 24 kHz speech or interleaved text and speech; it is a non-reasoning model that does not support enable_thinking. Its main weights hold 2,166,928,384 BF16 parameters, combining the LFM3 language model with the LFM3-VL vision encoder, the causal LFM3 audio encoder and a depth decoder for audio tokens, and the repository also ships the 73,559,090-parameter LiquidCodec tokenizer, for 2,240,487,474 in the released checkpoint. The card states a 32K-token context while the backbone configuration allows 128,000 positions, so no exact context is recorded; the tokenizer assigns 125,033 distinct text IDs.

Benchmark scores