← all releases
Liquid AI·Jan 6, 2026·v2.5·open source
LFM2.5
LFM2.5 is a new generation of small foundation models built on the LFM2 architecture, optimized for on-device and edge deployments. The model family includes variants such as LFM2.5-1.2B-Base, LFM2.5-1.2B-Instruct, and extends to Japanese, vision language, and audio language models. It is released as open weights on Hugging Face and exposed through the LEAP platform.
AI AgentsAgentic AILLMs
overview
LFM2.5 is a new generation of small foundation models built on the LFM2 architecture, optimized for on-device and edge deployments. The model family includes variants such as LFM2.5-1.2B-Base, LFM2.5-1.2B-Instruct, and extends to Japanese, vision language, and audio language models. It is released as open weights on Hugging Face and exposed through the LEAP platform.
key features
- 01Hybrid LFM2 architecture designed for fast and memory-efficient inference on CPUs and NPUs
- 02Extended pretraining from 10T to 28T tokens
- 03Supervised fine-tuning, preference alignment, and large-scale multi-stage reinforcement learning for instruction following, tool use, math, and knowledge reasoning
related productsbrowse all →
llmockopen sourcellmock is a deterministic mock LLM server designed for testing purposes. It provides a real HTTP server with authentic SSE streams, allowing developers to simulate interactions with various LLM APIs like OpenAI, Claude, and Gemini without incurring costs or dealing with non-deterministic results.AIBuildAIopen sourceAIBuildAI is an AI agent that autonomously constructs AI models. Given a specific task, it initiates an agent loop to analyze the problem, design models, and execute training processes, all without human intervention.MiroThinkeropen sourceMiroThinker is an open-source search agent model developed by MiroMindAI, designed for tool-augmented reasoning and real-world information seeking. It aims to match the deep research capabilities of leading AI models like OpenAI's Deep Research and Google's Gemini Deep Research.