Added on Oct 29,2026

LFM3 is the third generation of Liquid Foundation Models from Liquid AI, a family of small open-weight AI models designed to run on phones, laptops, robots and other edge devices. The models bring text, vision, audio, omni and decision capabilities onto one shared backbone.
The first release includes six models in the 2B parameter range, all branched from LFM3-1.7B-Base, which was pre-trained on 22 trillion tokens using a new hybrid architecture of short convolutions, sliding-window attention and global attention.
LFM3-1.7B is a text model for local agents with a 256K-token context window that fits in about 1 GB at 4-bit precision. It is post-trained for tool calling and agent , and can handle tasks such as inbox triage, calendar management and web research.
LFM3-VL-2B adds image and video understanding with step-by-step reasoning and spatial grounding. LFM3-Audio-2B supports voice chat, transcription, speech synthesis and translation across 18 languages, while LFM3-Omni-2B combines vision, audio and language in one model. The d1-2B and d1-Omni-2B decision models answer many questions about an input in a single pass, each with a probability.
Every model can be fine-tuned on a single GPU, and the release includes fine-tuning notebooks and examples covering industrial inspection and robotics.
It integrates with llama.cpp, MLX, vLLM, SGLang, ONNX and Ollama, and runs on Apple, AMD, NVIDIA, Arm, Intel and Qualcomm hardware. Models are also available through OpenRouter and Vercel.
Pricing: the base and post-trained weights are free to download on Hugging Face. The models can also be used through the Liquid AI API and Playground.
It is suited to developers, researchers and companies building private, on-device AI features, local agents, voice assistants and robotics applications.
Similar tools to compare with LFM3:
| Tool | Pricing | Best for |
|---|---|---|
| Pollinations | Freemium | Developers and hobby builders who want a single API for adding AI images, audio and text to their websites, bots and app... |
| NemoClaw | Free | Developers and platform teams who want to run always-on AI agents on NVIDIA hardware under policy-based security control... |
| Agnes AI | Freemium | Developers building AI-powered apps who want access to several models through one API. |
| DiagramGPT | Free | Developers and students who need to sketch system designs and workflows quickly by describing them in words. |
| OrcaRouter | Freemium | Developers and teams who want to balance AI quality and cost across many models. |
Leave a review for the community
No Review Found