LFM2.5 2.6B model competitive with 4x larger models
Summary
Liquid AI's LFM2.5-2.6B is a 2.6B-parameter model optimized for on-device deployment with a 128K context window. It claims competitive performance against larger models on tool use and agentic tasks, supported by agentic RL and multiple deployment options. The page details inference speed, memory footprint, and various formats (Transformers, vLLM, llama.cpp, etc.) and notes it's suitable for agentic workloads rather than coding tasks.