LFM3-Audio-2B

A 2B on-device voice model from Liquid AI: speech in and out, transcription, synthesis and translation in 18 languages.

In short. LFM3-Audio-2B is a small model that listens and talks. It can turn speech into text, read text aloud, translate speech, and answer spoken questions, all on a laptop or edge device without a cloud connection. Voice chat currently works in English only.

Summary

Liquid AI released LFM3-Audio-2B on 27 October 2026 as the audio member of its LFM3 family. It has about 2 billion parameters and takes text or 16 kHz audio as input. It answers with text, 24 kHz speech, or both interleaved, and it has a 32K-token context.

One model covers several jobs: voice chat with streaming replies, transcription, speech synthesis and speech translation across 18 languages, audio understanding (answering questions about speech, sounds and events in a recording), and turning spoken requests into tool calls. The model card says voice chat currently supports English only.

It uses a new in-house audio encoder and an audio tokenizer called LiquidCodec. The encoder is causal, so the model can start processing while you are still speaking; Liquid reports a time to first token as low as 50 milliseconds and speech generation at about 87 times real time on a single Apple M4 Max CPU thread. Liquid's own tests report an English word error rate of 6.2% for speech recognition. All of these figures are the developer's own.

Checkpoints come in BF16 (about 4 GB), GGUF for llama.cpp, MLX 4-bit for Apple silicon and ONNX. The license is the LFM Open License v1.0, which allows commercial use only below US$10 million of annual revenue.

Good for

  • Private, on-device voice assistants that can listen, reason, call tools and speak back
  • Transcription and speech translation without sending audio to the cloud
  • Text-to-speech in a small package on laptops and edge devices
  • Startups and small companies, since commercial use is free below US$10 million of revenue

Watch out for

  • Voice chat currently supports English only, even though transcription and synthesis cover 18 languages
  • Commercial use by entities at or above US$10 million of annual revenue is not licensed
  • It is a small model, and the card does not recommend it for agentic coding or complex multi-step reasoning
  • Speed and accuracy figures are Liquid's own measurements

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Memory figures for each variant of LFM3-Audio-2B. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.

Variants Parameters Download size Memory needed Fits on
LFM3-Audio-2B BF16 2B 4 GB 4.6 GB Laptop or phone

Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more.

The license in plain words

An Apache-2.0-style license from Liquid AI with one big catch. Commercial use is licensed only while your company, together with its affiliates, earns under US$10 million a year. Above that line you are not licensed and need your own agreement.

  • Free commercial use only below a revenue cap

More about LFM Open 1.0

See also

Frequently asked questions

What is LFM3-Audio-2B?
LFM3-Audio-2B is a small model that listens and talks. It can turn speech into text, read text aloud, translate speech, and answer spoken questions, all on a laptop or edge device without a cloud connection. Voice chat currently works in English only.
Can I use LFM3-Audio-2B commercially?
Yes, with conditions. LFM3-Audio-2B is released under the LFM Open 1.0, which allows commercial use but adds extra rules: Free commercial use only below a revenue cap. Read the license text before you ship a product.
Is LFM3-Audio-2B open source?
LFM3-Audio-2B is open-weight, not open source in the strict sense. You can download the model files, but the LFM Open 1.0 adds conditions that the Open Source Initiative definition does not allow.
Can I run LFM3-Audio-2B on my own computer?
Yes. The smallest variant, LFM3-Audio-2B, needs about 4.6 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was LFM3-Audio-2B released?
LFM3-Audio-2B was released by Liquid AI on October 27, 2026.
How long a context window does LFM3-Audio-2B have?
LFM3-Audio-2B supports a context window of up to 32K tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 8, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
LFM3 Liquid AI 1.73B – 2.17B 256K LFM Open 1.0 (Conditional) 1.19 GB Laptop or phone
Nemotron 3.5 Transcribe NVIDIA 34B 4.3B active 64K OpenMDW-1.1 (Permissive) 25 GB Gaming GPU
Parakeet TDT 0.6B v3 NVIDIA 627M – CC BY 4.0 (Permissive) 2.89 GB Laptop or phone
Kokoro hexgrad 82M – Apache 2.0 (Permissive) 0.38 GB Laptop or phone