Liquid Foundation ModelsUnclaimed

Liquid AI builds Liquid Foundation Models (LFMs), a family of AI models designed to run on the devices people already use as well as in the cloud. The company says its models are compute-optimized for low latency, privacy and specific hardware, so they can run on phones, laptops, cars, edge hardware and on-premises servers.

The site lists 65 LFMs shipped, more than 5,800 variants, and 61.0 million downloads. Models are available in common runtimes including llama.cpp, MLX, ONNX, Ollama, SGLang and vLLM, and the company names AMD, Qualcomm, Arm and NVIDIA as silicon partners. Its product lines include LFM2.5 and LFM3 language models, d1 and d1-ultra decision models, and audio and vision-language models.

The company is aimed at businesses building AI into their own products and devices. The site names automotive, consumer electronics, e-commerce, financial services, healthcare, industrial and robotics, and defence as target industries, and says Mercedes-Benz and Shopify work with it. For custom work, Liquid says it builds and optimizes models for a customer's deployment environment, provides OEM and on-premises support, and states that customer data never leaves the customer's environment and is not used to train models.

The open weight models are free under the LFM Open License for developers, researchers, non-profits and businesses with less than $10 million in annual revenue. Research, education and non-profit use is free with no revenue limit. Businesses above $10 million in annual revenue need a paid enterprise license, priced by deployment scale and sold through sales.

Hosted access is billed per token through the Liquid Console, OpenRouter or Vercel AI Gateway, with the same price on each platform. Prices vary by model; for example, LFM3-1.7B costs $0.08 per million input tokens and $0.20 per million output tokens. Every new organization gets $2 in free credit on Liquid Console.