Liquid Models

Independent guide to Liquid Foundation Models

Liquid Models for efficient AI anywhere.

Track the Liquid model landscape across LFM2, LFM2.5, text, vision-language, audio, and nano models built for edge, cloud, and hybrid deployment.

Explore models Official docs

Reference snapshot based on Liquid AI public model information. This site is independent and not affiliated with Liquid AI.

Abstract Liquid Models signal map
4 model groups
LFM2 core family
LFM2.5 latest family
3 deployment modes

What the keyword means

Liquid models combine small footprints with practical deployment targets.

Liquid AI describes LFMs as hybrid models rooted in dynamical systems and signal processing. For builders, the important idea is simpler: capable models that can be customized and deployed close to the work.

01

Device aware

Designed for smartphones, laptops, vehicles, robotics, and other environments where latency, memory, and privacy matter.

02

Multimodal

Model families cover text, image-text workflows, audio conversations, and specialized task models.

03

Customizable

LFMs are positioned for fine-tuning, task specialization, extraction, RAG, tool use, and enterprise deployment needs.

04

Hybrid ready

The same model strategy can span on-device inference, cloud scale, and private hybrid systems.

Model map

Meet the main Liquid model categories.

Use the filters to focus the map, then verify availability and licensing through official Liquid AI channels.

Showing 4 model groups

Text On-device / cloud / hybrid

Text Models

General language models for instruction following, retrieval workflows, extraction, and agentic tasks.

  • LFM2-350M
  • LFM2-700M
  • LFM2-8B-A1B
  • LFM2-24B-A2B
  • LFM2.5-1.2B-Base
  • LFM2.5-1.2B-Instruct
  • LFM2.5-1.2B-Thinking
  • LFM2.5-350M
Vision-language Multimodal

Vision-Language Models

Models that combine image and text understanding for low-latency visual intelligence and device-aware deployment.

  • LFM2-VL-450M
  • LFM2-VL-3B
  • LFM2.5-VL-450M
  • LFM2.5-VL-1.6B
Audio Conversation

Audio Model

End-to-end audio and text generation for responsive voice interactions with a compact foundation model footprint.

  • LFM2.5-Audio-1.5B
Nano Task specific

Nano Models

Tiny customized models for targeted jobs such as extraction, RAG, math, translation, PII workflows, and tool use.

  • LFM2-350M-Extract
  • LFM2-1.2B-RAG
  • LFM2-350M-Math
  • LFM2-ColBERT-350M
  • LFM2-1.2B-Tool
  • LFM2-350M-ENJP-MT

Deployment pattern

One model strategy, three runtime surfaces.

On device

Best for private data, offline resilience, low round-trip latency, and local hardware control.

Cloud

Best for centralized orchestration, elastic workloads, managed access, and shared inference services.

Hybrid

Best when sensitive work stays local while larger workflows coordinate with cloud services.

Benchmark snapshot

Small Liquid models are tracked across knowledge, instruction, math, and multilingual tests.

These figures summarize the public LFM2 benchmark table from Liquid AI. Always check the official page before making buying or architecture decisions.

LFM2 benchmark snapshot
Benchmark LFM2-350M LFM2-700M LFM2-1.2B
MMLU 43.43 49.90 55.23
GPQA 27.46 28.48 31.47
IFEval 65.12 72.23 74.89
GSM8K 30.10 46.40 58.30
MMMLU 37.99 43.28 46.73
Verify the latest official model page

Build with LFMs

Start from official resources when model access, licensing, or deployment details matter.

Docs Read implementation guides LEAP Customize and deploy Hugging Face Browse model releases Playground Try Liquid models

FAQ

Quick answers for Liquid Models searchers.

What are Liquid models?

In current AI search usage, Liquid models usually means Liquid AI's Liquid Foundation Models: efficient model families spanning text, vision-language, audio, and task-specific nano models.

What is the difference between LFM2 and LFM2.5?

LFM2 is the core family; LFM2.5 extends that line with newer text, vision-language, and audio models. Confirm exact availability through Liquid AI's official model page.

Can Liquid models run locally?

Liquid AI positions LFMs for CPU, GPU, and NPU deployment across on-device, cloud, and hybrid environments, depending on the model and licensing path.

Is this an official Liquid AI site?

No. Liquid Models is an independent educational resource. Product details, benchmarks, access terms, and licensing should be verified with Liquid AI.