# Liquid AI LFM3: specialized models that run on device

A summary of the LFM3 model family released by Liquid AI, based on its official announcement.

- Language: en
- Published: 2026-11-01
- Tags: automation, local-ai
- URL: https://devs7ud.io/en/trends/sensing-automation-2026-11-02-en
- Translation (ko): https://devs7ud.io/trends/sensing-automation-2026-11-02.md

## Specialized models that run on device

There is one AI automation item this time. Liquid AI released a family that groups text, vision, audio, omni and decision models.

## Specialized models that run on device

- [Liquid AI LFM3 announcement](https://www.liquid.ai/blog/lfm3) — The LFM3 model family

**Liquid AI LFM3 (October 27).** Liquid AI released LFM3, a family that groups text, vision, audio, omni and decision models. They are 2B-class open-weight models, with a path to fine-tune them and run them on device. I read it as a candidate for splitting out steps like sorting and tidying to a local model, instead of sending every step to a large cloud model. The performance figures are the vendor's own, so I did not treat them as independent results, and on-device latency and commercial terms need separate checks.

## View

I read it as a candidate for moving only steps like sorting and tidying to a local model, rather than handing every pipeline step to a large one. The performance figures are the vendor's own, so I will hold judgment until I measure it myself. That is my interpretation.

## Sources

- [Introducing LFM3](https://www.liquid.ai/blog/lfm3), Liquid AI