Liquid AI enables on-device AI with new language model for Raspberry Pi

AI startup Liquid AI, founded by former MIT computer scientists, has released LFM2.5-2.6B, an open-weight language model optimized for agentic workloads. The model is designed to operate entirely on local hardware, including smartphones, laptops, and devices as small as a Raspberry Pi, eliminating the need for cloud-based inference or dedicated GPUs.

This capability unlocks edge AI applications and provides greater flexibility for enterprises, particularly those in regulated industries or handling sensitive data that cannot be transmitted to the cloud. LFM2.5-2.6B is best suited for high-volume, well-defined agent tasks such as tool calling, document management, calendar and workflow automation, and background routines. It is also suitable for environments with limited connectivity, like vehicles and robotics.

The model and its base checkpoint (LFM2.5-2.6B-Base) are available on Hugging Face, supporting major inference stacks including llama.cpp and ONNX. This broad compatibility positions the model for deployment across consumer hardware, enterprise infrastructure, and embedded systems. Liquid AI has also released an open-source fine-tuning framework called LEAP.

Liquid AI positions LFM2.5-2.6B not as a competitor to large frontier models, but as a solution for enterprise applications where factors like latency, privacy, deployment flexibility, and inference costs are prioritized over benchmark leadership.