Liquid AI is an artificial intelligence company that develops Liquid Foundation Models (LFMs), a family of general-purpose models built on a novel architecture designed for efficiency across edge devices, AI PCs, and cloud infrastructure. The company, founded in 2023 and headquartered in the United States, grew out of research into alternative neural network architectures aimed at strong performance under tight latency, memory, and privacy constraints.
The company's core thesis is that many real-world deployments cannot rely on large cloud-hosted models because of cost, latency, connectivity, or data-privacy requirements. Liquid AI's models target on-device and embedded scenarios where inference must run locally and quickly, while still delivering competitive quality relative to larger models.
Liquid AI's product line has expanded with the LFM2 family, which the company describes as fast on-device foundation models built on a hybrid architecture, with subsequent releases extending to additional sizes, mixture-of-experts variants, and vision and audio language capabilities. Many of these models have been published openly on Hugging Face under permissive licensing for research and academic use.
The company has raised significant venture funding, with a Series A round reported in late 2024, and has cited enterprise engagements with large organizations to bring low-latency foundation-model inference into real-world products and devices. Its positioning emphasizes running capable models close to where data is generated.
Liquid AI operates in a competitive landscape that includes other efficient and small-model providers as well as the on-device offerings of major foundation-model labs. Its differentiation centers on architecture-level efficiency claims and a focus on edge and embedded deployment rather than only large cloud models.
The company is best suited to teams building AI features for devices, applications, or environments where local inference, predictable latency, and data privacy matter more than access to the very largest cloud models.