Liquid AI announced on October 27, 2026, the LFM3 family, a new generation of Liquid Foundation Models with six open-weight models in the 2-billion-parameter range. All branch from a shared LFM3-1.7B-Base backbone trained on 22 trillion tokens.
The lineup includes LFM3-1.7B (text, 256K context), LFM3-VL-2B (vision with video support), LFM3-Audio-2B (18 languages, speech in and out), LFM3-Omni-2B (full multimodal) and two decision models (d1-2B and d1-Omni-2B) that return calibrated answers in a single forward pass.
The tri-hybrid architecture combines short convolutions with sliding-window attention and global attention, prioritizing efficiency for on-device use. The text model fits in about 1 GB at 4-bit quantization and can be fine-tuned on a single GPU. Weights are available on Hugging Face under the LFM Open License v1.0, with free commercial use for companies under $10 million in annual revenue.
The company highlights gains in agentic tasks and tool-calling versus the previous LFM2 generation, while maintaining high speed on edge hardware, including day-0 support on AMD, Qualcomm, NVIDIA and other platforms.
The release reinforces the trend of compact, open-weight multimodal models aimed at local and specialized execution, reducing reliance on large cloud models.
Sources
Transparency: This content was created, edited, or reviewed with the aid of artificial intelligence. Information was cross-referenced with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot
.avif)