API
< Back to situations

Monitor this situation.

[SITUATION] · [QUIET] · [TECHNOLOGY]

2 clusters · 8 sources · 14 days · First seen · Last updated

Liquid AI open-weight model releases

Overview

Liquid AI has expanded its offerings of open-weight models designed for local deployment and edge computing.

In mid-October, the company released LFM2.5-24B-A2B, a reasoning model utilizing a mixture-of-experts architecture. This model activates 2 billion parameters per token to reduce computational requirements while maintaining a 131,000-token context window. It was optimized for agentic tasks and demonstrated high speeds on Apple’s M5 Max chip.

By late October, Liquid AI launched the LFM3 family, consisting of six multimodal models based on the LFM3-1.7B-Base foundation. These models are approximately 2 billion parameters each and include specialized variants for text, vision-language, audio, decision-making, and an “omni” model for simultaneous processing of text, images, video, and audio. The LFM3 text model features an expanded 256K-token context window, and the architecture employs a tri-hybrid approach combining gated short convolutions, sliding-window attention, and global attention.

Entities

Liquid AI · Hugging Face · Qwen · Apple

Timeline

  1. [TECHNOLOGY] 8 sources
    Liquid AI launches LFM3 family of small, open-weight models

    Liquid AI has unveiled LFM3, a family of six 2B-parameter open-weight models designed for local, efficient deployment on consumer hardware.

  2. [TECHNOLOGY] 2 sources
    Liquid AI releases LFM2.5-24B-A2B open-weight model for local hardware

    Liquid AI released LFM2.5-24B-A2B, an open-weight 24B-parameter model optimized for local use on consumer hardware via a mixture-of-experts architecture.

Sources

ascii.jp · cryptobriefing.com · dev.to · gigazine.net · k-tai.watch.impress.co.jp · pc.watch.impress.co.jp · smartphonology.it · weekly.ascii.jp