230M (LFM2 hybrid architecture; 14 layers: 8 double-gated LIV conv + 6 GQA)
32,768 tokens
LFM Open License v1.0
Open-weight 230M instruction-tuned LFM2.5 release positioned by Liquid AI as its smallest model yet for edge and on-device agentic workflows. Official release materials describe LFM2.5-230M as a text-only hybrid LFM2 model distilled from LFM2.5-350M and then post-trained with a lightweight three-stage recipe: supervised fine-tuning, direct preference optimization, and multi-domain reinforcement learning. The Hugging Face model card lists 19T pretraining tokens, a 32,768-token context window, 65,536 vocabulary size, and support for ten languages including English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, and Spanish. Liquid emphasizes very fast low-footprint inference across constrained hardware, reporting 213 tok/s decode on a Galaxy S25 Ultra and 42 tok/s on a Raspberry Pi 5, with day-one support across Transformers, vLLM, SGLang, llama.cpp, MLX, and ONNX companion formats. Benchmark tables in the release compare the model against LFM2.5-350M, prior LFM2-350M, Granite 4.0 350M variants, Qwen3.5-0.8B Instruct, and Gemma 3 1B IT, with especially strong positioning on tool use, function calling, and data extraction tasks such as BFCLv3, BFCLv4, and CaseReportBench. Release guidance recommends it for lightweight on-device agentic pipelines and structured extraction rather than reasoning-heavy math, code generation, or creative writing.