Reported information
Liquid AI releases LFM3, six compact models for local deployment
Liquid AI introduced LFM3 on 27 October 2026, a family of six downloadable-weight models of roughly two billion parameters. They cover text, images, audio and specialized decision tasks.
The variants share one backbone. Liquid AI says they can be customized on a single graphics processor and deployed on company-controlled devices or servers. The text model has a 256,000-token context window.
Webdiga's initial independent tests demonstrate local text and audio execution but find runtime and operating-system support still varies. Advertised performance needs assessment for each use case. Downloadable weights do not imply unconditional commercial licensing: Liquid lists free use until the user's company exceeds $10 million in annual revenue.