No cloud, no GPUs, no problem: Liquid AI's new model LFM2.5-2.6B brings powerful AI agents to devices as small as a Raspberry Pi

Liquid AI, a startup founded in 2023 by former MIT computer scientists, has unveiled a new open-weight language model named LFM2.5-2.6B. This model is specifically designed for agentic workloads and can run on local hardware, including smartphones, laptops, and even a Raspberry Pi, without needing cloud inference or GPUs. This makes it particularly suitable for high-volume, well-defined tasks that can be performed locally, such as tool calling, document management, calendar and workflow automation, and always-on background routines.

Additionally, it can be useful in connectivity-limited environments like vehicles and robotics. The model contains 2.6 billion parameters and supports a 128,000-token context window, and it includes native tool calling capabilities. Liquid AI offers both a pre-trained model and a base checkpoint for developers who wish to fine-tune it on Hugging Face, with support for major inference stacks like llama.cpp, MLX, vLLM, SGLang, and ONNX.

The company emphasizes that LFM2.5-2.6B is tailored for enterprises where latency, privacy, deployment flexibility, or inference costs are more critical than absolute benchmark performance. Maxime Labonne, Liquid AI's head of post-training, noted that the model was developed with the assumption that language models will increasingly be consumed through agent frameworks rather than traditional conversational interfaces, and that it is particularly good at using tools.