Liquid AI

What is Liquid AI?

Liquid AI develops device-native foundation models designed to run locally on hardware outside data centers, including phones, laptops, and cars. The models address latency, privacy, and hardware constraints by keeping data on the device. Access is available through an API or local deployment using supported runtimes.

What are the features of Liquid AI?

• On-device multimodal models

  LFM3 supports multiple modalities for specialized tasks and runs local agents in approximately 1GB of memory.
• Decision models

  d1-ultra and Open d1 process text, vision, and audio inputs to automate decisions, with Open d1 operating in under 50 ms.
• Vision-language processing

  LFM2.5-VL-3B analyzes screen content and other visual inputs directly on edge devices.
• Runtime compatibility

  Models deploy via llama.cpp, MLX, ONNX, Ollama, SGLang, vLLM, and other runtimes across AMD, Qualcomm, Arm, and NVIDIA silicon.
• Quantization-aware distillation

  LFM2.5 Q4_0 optimizes model weights for edge deployment.

What are the use cases of Liquid AI?

• In-car AI

  Mercedes-Benz uses Liquid AI models to advance private, on-device speech, language understanding, and reasoning for automotive experiences.
• E-commerce automation

  Shopify deploys the models for scaled production decision-making.
• Local Mac applications

  MacPaw partners with Liquid AI to deliver on-device AI to Mac users without sending data off the machine.