Liquid AI's LFM2.5-2.6B runs agents on a Raspberry Pi with no GPU

Liquid AI released LFM2.5-2.6B, a 2.6B-parameter open-weight model built for agentic workloads that runs entirely on local hardware, from smartphones and laptops down to a Raspberry Pi. It supports a 128,000-token context window and native tool calling, with both the post-trained model and a base checkpoint on Hugging Face plus day-one support for major inference stacks. The target is high-volume, well-defined local tasks: tool calling, document management, calendar and workflow automation, and always-on background routines, plus connectivity-limited settings like vehicles and robotics. Coding-heavy work still needs larger models. The custom open weights license, like Kimi K3's, warrants a close read from enterprise legal teams.

View full digest for August 7, 2026