Liquid AI releases six open-weight LFM3 models on a shared backbone
The LFM3 family brings text, vision, video, audio, omni and decision workloads into six 2B-class models that can each be customized on a single GPU.
AT A GLANCE
Key points
- Six 2B-class models cover text, vision, video, audio, omni and decision workloads.
- The text model offers a 256K-token context window and fits in about 1 GB at 4-bit precision.
- All LFM3 models are open-weight and available on Hugging Face.
What is changing?
Liquid AI released the LFM3 family as open-weight models on Hugging Face on October 27. The lineup contains six 2B-class models derived from the LFM3-1.7B base: the LFM3-1.7B text model, LFM3-VL-2B with image and video input, LFM3-Audio-2B, LFM3-Omni-2B, and the d1-2B and d1-omni-2B decision models.
The company says its new tri-hybrid backbone combines short convolutions, sliding-window attention and global attention. The text model offers a 256K-token context window and fits in about 1 GB at 4-bit precision. The vision model accepts video, while the audio and omni models can produce text and speech. Every model is designed to be fine-tuned on one GPU.
Open weights can support local agents and domain-specific models, but the term does not mean that all training data and code are open source. The published performance comparisons are Liquid AI measurements, so teams should test latency, memory use and quality on their own hardware.
Explore the original sources
Source date: 27 October 2026.
Published by AI Radar: 27 Oct 2026, 22:05 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…