Liquid AI Unveils Revolutionary Non-Transformer Models with Superior Efficiency

The new Liquid Foundation Models promise state-of-the-art performance with reduced memory usage, challenging existing AI architectures.

Liquid AI, a startup founded by MIT researchers, has introduced Liquid Foundation Models (LFMs) that outperform traditional transformer-based AI models.
The LFMs are available in three sizes: 1.3B, 3.1B, and 40.3B parameters, with the largest being a Mixture of Experts model.
These models require significantly less memory, making them cost-effective and energy-efficient compared to competitors like Meta's Llama and Microsoft's Phi models.
Liquid AI's architecture is built from first principles, focusing on computational units to maximize knowledge capacity and reasoning.
The models are currently accessible through Liquid's platforms and are optimized for a variety of hardware, including Nvidia, AMD, Qualcomm, and Apple.