Liquid AI LFM2.5-350M: Agentic Tool Use at 350M Parameters, 95%+ Accuracy, <500MB Quantized

Liquid AI·medium signal

Liquid AI released LFM2.5-350M on March 31 — a 350M parameter model trained on 28T tokens with scaled RL, purpose-built for tool calling and data extraction where models this small typically fail. Partners report 95%+ tool-calling accuracy across multi-turn interactions in smart home, banking, and terminal use cases. Processes 40.4K output tokens/second on a single H100, fits under 500MB quantized. IFEval score of 76.96 makes it viable for edge-deployed agentic loops on constrained hardware.