LFM (Liquid Foundation Models)
Family overview — summarizes releases; licenses and availability belong to each release.
LFM is Liquid AI's series of hybrid language and multimodal models for on-device use, built from gated short-convolution blocks and a small number of grouped-query attention blocks. LFM2 (from July 2025) introduced open-weight dense models from 350M parameters upward and later an 8.3B mixture-of-experts model. LFM2.5 (from January 2026) extended pretraining and scaled up reinforcement-learning post-training; 2026 releases include LFM2.5-8B-A1B (May), LFM2.5-2.6B (August), and LFM2.5-VL-3B (August), alongside vision, audio, and encoder variants.43516710
- Website: Liquid AI models page (external site: liquid.ai)
- Documentation: LFM documentation (external site: docs.liquid.ai)
- Paper: LFM2 technical report (arXiv 2511.23404) (external site: arxiv.org)
- License: LFM Open License (external site: liquid.ai)
Releases assessed
| Release | Weights | Tier (USASI rubric v0.1) | License | Released |
|---|---|---|---|---|
| LFM2.5-2.6B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | LFM Open License v1.0 | Aug 4, 2026 |
| LFM2.5-8B-A1B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | LFM Open License v1.0 | May 28, 2026 |
What it is useful for
Liquid AI positions the models for on-device and other resource-constrained deployment, and its model cards recommend the current text models for agentic workflows, tool use, data extraction, retrieval-augmented generation, and structured outputs rather than knowledge-heavy or heavy coding tasks.289
Organization context
Provenance and derivatives
The LFM language models are pretrained by Liquid AI itself; the current text instruction and reasoning models are post-trained from Liquid's own base checkpoints (for example LFM2.5-2.6B-Base and LFM2.5-8B-A1B-Base). The vision-language models are not text-only LFMs: the LFM2.5-VL-3B model card says it combines the LFM2.5-2.6B language backbone with a SigLIP2 NaFlex vision encoder. Liquid's LFM2 announcement says an earlier Liquid model, LFM1-7B, was used as a teacher for knowledge distillation during LFM2 training.89114
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.