Skip to main content

Chat & Rewriting

Conversational interactions, text transformations, and summarization.

Tool Calling

Function invocation, agent workflows, and API orchestration.

Structured Output

JSON generation, form filling, and data extraction from text.

Classification

Intent detection, routing, content labeling, and triage.

LFM3 Models   Latest release

LFM3 is built on a new tri-hybrid architecture of gated short convolutions, sliding-window attention, and global attention, pre-trained on 22T tokens. It is far more expressive than LFM2 while maintaining its efficiency.

LFM3-1.7B

1.7B RecommendedText model for local agents and specialized applications, with 256K context and native tool calling.

LFM2.5 Models

LFM2.5 builds on the LFM2 architecture with extended pre-training and reinforcement learning for improved chat, instruction-following, and tool-calling performance.

LFM2.5-1.2B-Instruct

1.2B ChatInstruction-tuned for chat and general-purpose tasks.

LFM2.5-1.2B-Thinking

1.2B ReasoningOptimized for math and logical problem-solving.

LFM2.5-1.2B-JP-202606

1.2B JapaneseJapanese chat model for Japanese and English assistants, tool use, and structured outputs.

LFM2.5-24B-A2B

24B · 2B active MoEOur largest model. Reasoning model with 131K context, built for fast, private agents on-device.

LFM2.5-8B-A1B

8B · 1.5B active MoEMixture-of-experts reasoning model with 128K context for on-device tool calling and agentic tasks.

LFM2.5-350M

350M CompactCompact LFM2.5 model for edge devices and low latency deployments, under 1 GB of memory.

LFM2.5-230M

230M SmallestSmallest LFM2.5 model. Built for data extraction and lightweight on-device agents.

Examples

Explore practical implementations using text models:

Product Slogan Generator

Generate creative marketing slogans on-device using LFM2.5-1.2B-Instruct with single-turn generation and traditional Android Views.Platform: Android · Uses: LFM2.5-1.2B-Instruct

Web Content Summarizer

Share web pages from browsers to generate private, on-device summaries with web scraping and LFM2.5-350M processing.Platform: Android · Uses: LFM2.5-350M

Recipe Generator with Constrained Output

Generate structured JSON recipes using LFM2.5-1.2B-Instruct with llama.cpp JSON-schema constraints.Platform: Android · Uses: LFM2.5-1.2B-Instruct

Tool-Calling Agent

Build agents with tool invocation using LFM2.5-1.2B-Instruct and llama.cpp on Android.Platform: Android · Uses: LFM2.5-1.2B-Instruct

English-Korean Translation CLI

Bidirectional translation with automatic language detection. Fine-tuned LFM2-1.2B outperforms models 3x larger.Platform: Desktop · Uses: LFM2-1.2B (fine-tuned)