LFM2.5-2.6B Enables True Edge AI on Raspberry Pi and Phones

Why LFM2.5-2.6B is reshaping edge AI

Liquid AI’s freshly released LFM2.5-2.6B model can run on devices ranging from smartphones to a Raspberry Pi without cloud or GPU support, opening a new frontier for privacy‑first, low‑latency agents. The 2.6 billion‑parameter engine supports a 128,000‑token context window and native tool calling, making it ideal for document handling, calendar automation and background routines. Enterprises handling regulated data see immediate cost savings – the model consumes only electricity, with ↑ 220 tokens/sec on an Apple M5 Max. Reuters notes a surge in demand for on‑device AI.

Edge performance on minimal hardware

Benchmarks show roughly ↑ 113 tokens/sec on an AMD Ryzen AI Max+ 395 and about 30 tokens per second on a typical smartphone. Liquid’s own pandemic-era remote‑work shift accelerated interest in such locally hosted assistants. The company also released a mobile app, Apollo, allowing users to test the model on‑device.

“We designed LFM2.5-2.6B for agents that run independently, not just chatbots,” said Maxime Labonne, head of post‑training.

For heavy‑duty workloads, a single Nvidia H100 can push nearly 15,000 output tokens per second, equating to about 1.3 billion tokens per day – figures that remain vendor‑claimed. The model is available on Hugging Face with day‑one support for llama.cpp, MLX, vLLM, SGLang and ONNX, and can be fine‑tuned via the open‑source LEAP framework.

Licensing and enterprise outlook

LFM2.5-2.6B ships under the LFM Open License v1.0, free for organizations under $10 million revenue; larger firms must negotiate a commercial deal. This hybrid approach aims to fund continued research while keeping barriers low for startups. Partnerships, such as with MacPaw for an on‑device macOS assistant, illustrate immediate commercial traction.

Analysis by Dr. Aris Thorne (Artificial Intelligence Researcher).