Text Models

Liquid’s LFM text models range from 350M to 8B parameters, delivering ultra-low-latency generation while matching the performance of much larger models. They come in both dense and MoE variants to deploy flexibly across different devices.

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.

LFM2.5 Models   Latest release

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 · Recommended

Instruction-tuned for chat. Best for most use cases.

LFM2.5-1.2B-Thinking

1.2B · Reasoning

Optimized for math and logical problem-solving.

LFM2.5-1.2B-Base

1.2B · Pre-trained

Base model for finetuning or custom checkpoints.

LFM2.5-1.2B-JP

1.2B · Japanese

Fine-tuned model for high-quality Japanese text generation.

LFM2 Models

LFM2-8B-A1B

8B · 1.5B active · MoE

Mixture-of-experts model for on-device speed and quality.

LFM2-2.6B

2.6B

Highly capable model for deployment on most phones and laptops.

LFM2-2.6B-Exp

2.6B

RL-only post-trained checkpoint for improved math and reasoning.

LFM2-1.2B

1.2B · Deprecated

Use the new LFM2.5-1.2B-Instruct checkpoint instead.

LFM2-700M

700M

Mid sized model for deploying on most devices.

LFM2-350M

350M · Fastest

Our smallest model for edge devices and low latency deployments.