Liquid AI's LFM2.5-2.6B runs agentic workloads on a Raspberry Pi, with no cloud and no GPU
A 2.6-billion-parameter open-weight model designed specifically for agentic use, running entirely on local hardware from smartphones and laptops down to a Raspberry Pi.
Designed for agentic workloads is the distinguishing claim. Small models have run locally for years; what they have generally not done is tool use, structured output and multi-step planning reliably enough to drive an agent loop. Targeting that specifically is a different objective from compressing a chat model.
The Raspberry Pi figure matters because of what it rules out. No cloud means no per-token cost, no network dependency, no data leaving the device, and no vendor able to deprecate the model underneath a product.
The honest limit is capability. A 2.6B model is not competing with a 117B open-weight release on reasoning. It is competing with having nothing on the device at all, and that is a comparison it wins.