Liquid's 2.6B-parameter model runs on Raspberry Pi, no cloud or GPU required
Liquid, an AI startup founded in 2023 by former MIT computer scientists, has released LFM2.5-2.6B, an open-weight language model designed to run entirely on local hardware — from smartphones and laptops down to a Raspberry Pi — without relying on cloud inference or GPUs.
According to the company's researchers, the model is best suited for high-volume, well-defined agentic tasks executed locally: tool calling, document management, calendar and workflow automation, and always-on background routines. It also fits connectivity-limited environments like vehicles and robotics. Coding-heavy work, however, is better left to larger models.
For enterprises in regulated industries or those handling sensitive data that cannot be sent to the cloud, the model unlocks edge AI applications. Even for businesses without such concerns, running performant, task-specific agents at essentially the cost of electricity may prove appealing.
LFM2.5-2.6B has 2.6 billion parameters and supports a 128,000-token context window, including native tool calling. The name reflects the model generation (2.5) and parameter count (2.6B). Both the post-trained model and a base checkpoint, LFM2.5-2.6B-Base for fine-tuning, are available on Hugging Face.
Day-one support covers major inference stacks: llama.cpp, MLX, vLLM, SGLang, and ONNX — positioning the model for deployment across consumer hardware, enterprise infrastructure, and embedded systems. Liquid also offers an open-source fine-tuning framework called LEAP.
The company does not position LFM2.5-2.6B as a competitor to the largest frontier models, focusing instead on specialized local scenarios. Still, the custom open-weights license, similar to Moonshot's recently released Kimi K3, warrants close review by enterprise legal teams.
