---
title: "Liquid AI | Device-native foundation models."
description: "Liquid AI builds efficient Liquid Foundation Models (LFMs) for on-device, edge, and cloud AI with low latency, privacy, and hardware-aware deployment."
canonical: "https://www.liquid.ai/"
locale: "en"
---

# _Device_ _\-native_ foundation models.

Advanced intelligence for processors outside of data centers. Built for the latency, privacy, and hardware constraints of the physical world.

Models

[65 LFMs shipped](https://www.liquid.ai/models)

[62.6M downloads](https://huggingface.co/LiquidAI)

[6000+ variants](https://huggingface.co/LiquidAI)

Deployments

[Phones](https://www.liquid.ai/press/liquid-ai-to-power-on-device-intelligence-in-sharps-bestselling-mobile-phone), [laptops](https://www.liquid.ai/blog/lfm2-24b-a2b-from-cloud-to-ai-pc), [cars](https://www.wsj.com/cio-journal/mercedes-gives-its-cars-an-ai-upgrade-607daae2), [space](https://payloadspace.com/dphis-second-demo-brings-compute-space-to-orbit/), [e-commerce](https://www.liquid.ai/blog/liquid-ai-announces-multi-year-partnership-with-shopify-to-bring-sub-20ms-foundation-models-to-core-commerce-experiences), [on-prem](https://www.liquid.ai/e-commerce), [bio](https://www.liquid.ai/blog/longevitybench), [defense](https://www.liquid.ai/defense)

Runtimes

[llama.cpp](https://docs.liquid.ai/deployment/on-device/llama-cpp), [MLX](https://docs.liquid.ai/deployment/on-device/mlx), [ONNX](https://docs.liquid.ai/deployment/on-device/onnx), [Ollama](https://docs.liquid.ai/deployment/on-device/ollama), [SGLang](https://docs.liquid.ai/deployment/gpu-inference/sglang), [vLLM](https://docs.liquid.ai/deployment/gpu-inference/vllm)

& more

Silicon partners

AMD, Qualcomm, Arm, NVIDIA

& more

[Explore Liquid Foundation Models](https://www.liquid.ai/models) [Connect with us](https://www.liquid.ai/connect)

## Latest news

[Read more](https://www.liquid.ai/news/models)

- [11.12Engineering — LiquidCodec: High-Fidelity Audio Tokenization Optimized for Speed](https://www.liquid.ai/blog/liquidcodec)
- [10.27Models — Introducing LFM3: Every Modality for Specialized Intelligence](https://www.liquid.ai/blog/lfm3)
- [10.22News — RIPL: Real-time Intelligence When Every Millisecond Counts](https://www.liquid.ai/blog/low-latency)
- [10.15Models — d1-ultra: Introducing the World’s Best Decision Model](https://www.liquid.ai/blog/d1-ultra)
- [10.13Models — LFM2.5-24B-A2B: Built for Fast On-Device Agents](https://www.liquid.ai/blog/lfm2-5-24b-a2b)
- [10.07Models — Open d1: Edge decision models for text, vision, and audio](https://www.liquid.ai/blog/d1-open)
- [10.05Models — Introducing d1: The most capable decision model, now with vision](https://www.liquid.ai/blog/d1-decision-model)
- [09.24Models — LFM2.5-VL-DSpark: Accelerating vision-language models on edge and beyond](https://www.liquid.ai/blog/lfm2-5-vl-dspark)
- [09.23News — Liquid AI, in Collaboration with Qualcomm Technologies, Brings Personal AI Context to Devices Powered by Snapdragon](https://www.liquid.ai/blog/liquid-context-qualcomm)
- [09.17Engineering — Setting the Frontier of Aging Biology with Liquid Foundation Models ](https://www.liquid.ai/blog/longevitybench)
- [08.24News — Introducing Pipette: A benchmarking suite for on-device intelligence](https://www.liquid.ai/blog/pipette-on-device-ai-benchmarking-by-liquid-ai)
- [08.20Models — LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook](https://www.liquid.ai/blog/lfm2.5-dspark)
- [08.19Models — LFM2.5 Q4\_0: Quantization-Aware Distillation for Edge Deployment](https://www.liquid.ai/blog/qad)
- [08.18Engineering — Designing Loops for Production-Grade Work](https://www.liquid.ai/blog/agent-loops)
- [08.12Models — LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge](https://www.liquid.ai/blog/lfm2-5-vl-3b)

AnnouncementNOV 12, 2026

Engineering

## LiquidCodec: High-Fidelity Audio Tokenization Optimized for Speed

[Read the announcement](https://www.liquid.ai/blog/liquidcodec)

> “Liquid AI's models are the best for the scaled production deployment - pareto-optimal, beating much larger rivals. That's why we use them at Shopify.”

Mikhail Parakhin, CTO, Shopify

## Liquid AI is an efficiency‑first foundation model company.

Intelligence should be ubiquitous, **co-designed with hardware**, and never static.

Demand for intelligence compounds exponentially, but cloud-only serving can’t keep up on cost, power, or latency. Meanwhile, 2.2 billion devices that can host a foundation model ship every year. We co-design our models with that hardware from first principles, and build them to keep improving after they ship.

[Read our thesis](https://www.liquid.ai/company#thesis)

## Designed with the hardware in the loop

We score candidate architectures on task quality and on latency and memory measured on the target device, then optimize the models to fit its budget.

- [Recursively improved on real hardware — LFM3’s architecture was chosen by searching across 14 device and backend combinations, scoring each candidate on task quality and on latency and memory measured on the device. — How we design for hardware](https://www.liquid.ai/automated-foundation-model-design)
- [Measured on device — Pipette is our open-source benchmark, built with Artificial Analysis, with lab-verified results for more than 1,000 model, quantization, runtime, and device configurations. — Explore Pipette](https://pipette.liquid.ai/)
- [Architectures that evolve — Our AFMD algorithms evolve model architectures for a target’s tasks, metrics, and hardware, producing designs with smaller caches and fewer parameters. — Read about AFMD](https://www.liquid.ai/blog/automated-architecture-synthesis-via-targeted-evolution)

## Featured Partners

- [Alef Education: Alef Education Collaborates with Liquid AI to Advance AI in Education Globally](https://www.liquid.ai/press/alef-education-collaborates-with-liquid-ai-to-advance-ai-in-education-globally)
[AMD: From Cloud to Keyboard: LEAP Expands to Laptops with Cutting-Edge AMD Ryzen™ Power](https://www.liquid.ai/blog/liquids-edge-ai-platform-leap-expands-support-to-laptops-with-best-in-class-performance-on-amd-ryzen-tm-and-ryzen-ai-tm-processors)
[Arm: Introducing LFM3: Every Modality for Specialized Intelligence](https://www.liquid.ai/blog/lfm3)
[Baseten: Deploy LFMs on Baseten](https://docs.liquid.ai/deployment/gpu-inference/baseten)
[Capgemini: Liquid AI announces collaboration with Capgemini to build next-generation AI solutions for enterprises](https://www.liquid.ai/blog/liquid-ai-announces-collaboration-with-capgemini-to-build-next-generation-ai-solutions-for-enterprises)
[DPhi Space: Fine-tune a vision-language model on satellite imagery](https://docs.liquid.ai/examples/customize-models/satellite-vlm)
[G42: G42 and Liquid AI Partner to Deliver Private, Local and Efficient AI Solutions to Enterprises at Scale](https://www.liquid.ai/press/g42-and-liquid-ai-partner-to-deliver-private-local-and-efficient-ai-solutions-to-enterprises-at-scale)
[Generative Bionics: Humanoid robots powered by Physical AI](https://gbionics.ai/)
- [Hugging Face: Download open LFM weights](https://huggingface.co/LiquidAI)
[Insilico Medicine: Liquid AI and Insilico Medicine Announce Strategic Partnership Delivering Lightweight Scientific Foundation Models for Drug Discovery](https://www.liquid.ai/press/liquid-ai-insilico-medicine-partnership)
[llama.cpp: Run LFMs locally with Llama, the desktop app from the llama.cpp team](https://docs.liquid.ai/deployment/on-device/llama)
[LM Studio: From Cloud to AI PC: Launching Liquid’s Largest LFM2 Model Alongside Our Growing Partner Ecosystem](https://www.liquid.ai/blog/lfm2-24b-a2b-from-cloud-to-ai-pc)
[MacPaw: MacPaw Partners with Liquid AI to Bring On-Device AI to Millions of Mac Users](https://www.liquid.ai/blog/macpaw-partners-liquid-ai-on-device-ai-mac-users)
[Mercedes-Benz: Liquid AI and Mercedes-Benz partner to scale embedded in-car intelligence](https://www.liquid.ai/press/liquid-ai-and-mercedes-benz-partner-to-scale-embedded-in-car-intelligence)
[Modal: From Cloud to AI PC: Launching Liquid’s Largest LFM2 Model Alongside Our Growing Partner Ecosystem](https://www.liquid.ai/blog/lfm2-24b-a2b-from-cloud-to-ai-pc)
[Ollama: Run LFMs locally with Ollama](https://docs.liquid.ai/deployment/on-device/ollama)
- [OpenRouter: Introducing d1: The most capable decision model, now with vision](https://www.liquid.ai/blog/d1-decision-model)
[Qualcomm: Liquid AI, in Collaboration with Qualcomm Technologies, Brings Personal AI Context to Devices Powered by Snapdragon](https://www.liquid.ai/blog/liquid-context-qualcomm)
[QuantumBlack: The case for Liquid Foundation Models](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-case-for-liquid-foundation-models)
[Rain Technology Solutions: Digitally empowering medics to preserve life in chaos](https://raintechsolutions.com/)
[SGLang: Serve LFMs with SGLang](https://docs.liquid.ai/deployment/gpu-inference/sglang#sglang)
[Shopify: Liquid AI Announces Multi‑Year Partnership with Shopify to Bring Sub‑20ms Foundation Models to Core Commerce Experiences](https://www.liquid.ai/blog/liquid-ai-announces-multi-year-partnership-with-shopify-to-bring-sub-20ms-foundation-models-to-core-commerce-experiences)
[Unsloth: Run and fine-tune LFM2.5 with Unsloth](https://unsloth.ai/docs/models/tutorials/lfm2.5)
[Vercel: Introducing d1: The most capable decision model, now with vision](https://www.liquid.ai/blog/d1-decision-model)

## Tell us what you’re building, we’ll help you ship it.

From picking the right model to customizing and deploying it on your hardware, our team works with you end to end.

[Connect with us](https://www.liquid.ai/connect)
