Promotional spot · Feature your product at the top bar.Get spotlight

Liquid

Liquid builds device-native foundation models and decision models designed for the latency, hardware, and deployment constraints of running AI outside data centers.

Overview

Liquid is an efficiency-first foundation model company that builds device-native foundation models (LFMs) and decision models for running advanced intelligence on processors outside of data centers. Its models are designed around the latency, hardware, and deployment constraints of the physical world, addressing the gap between growing demand for AI and the cost, power, and latency limits of cloud-only serving.

Key Features

  • Liquid Foundation Models (LFMs): A family of models with 65 LFMs shipped and 5,800+ variants, spanning text, vision, audio, and decision capabilities.
  • On-device and edge focus: Models are built to run on phones, laptops, cars, and other devices, with frontier AI sized to run in roughly 1GB and local agents running in about 1GB.
  • Decision models (d1): Purpose-built for fast decisions, with open d1 models delivering decisions in under 50 ms and d1-ultra positioned for decision tasks.
  • Runtime and silicon compatibility: Support across runtimes including llama.cpp, MLX, ONNX, Ollama, SGLang, and vLLM, with silicon partners including AMD, Qualcomm, Arm, and NVIDIA.
  • API access: Developers can get an API key to automate decisions with AI using Liquid's models.

Typical Use Cases

  • On-device agents: Run local agents directly on hardware with a small footprint, suited to devices where cloud round-trips are impractical.
  • Real-time decision making: Use decision models such as d1 for applications where millisecond-level response times matter.
  • Embedded and physical AI deployments: Deploy models across environments listed by Liquid, including phones, laptops, cars, space, e-commerce, on-prem, bio, and defense.

How It Works

Liquid co-designs its models with target hardware from first principles and builds them to keep improving after they ship. Developers can explore the LFM model family and obtain an API key to start automating decisions with AI. Models are deployable across a range of runtimes and silicon partners, supporting both local and server-based deployment.

Who It's For

Liquid suits developers, engineering teams, and organizations building AI into devices and physical systems where latency, hardware limits, and local execution matter. It is relevant to teams working on on-device agents, embedded decision-making, and deployments that cannot rely on data-center-only serving.

Liquid provides compact, hardware-aware foundation and decision models for teams bringing AI to devices and edge environments.

Explore Similar Tools

Unassisted coding diagnostic and practice tool for JavaScript, TypeScript, and Python without autocomplete or AI assistance.

AI inbound voice receptionist handling 24/7 business phone calls, scheduling appointments, answering FAQs, and escalating urgent issues.

AI remembrance and voice preservation web app that lets you talk with the people you miss again through text or synthesized voice.

Turn MP3, WAV, and M4A recordings into editable MIDI online, using AI to transcribe clear melodies, vocals, piano phrases, and guitar riffs.

Nano Banana CN is a Chinese-friendly AI image and video studio. An API is available for image models. Nano Banana CN is an independent site and is not affiliated with Google.

Turn one photo into a trending AI video: Bulin 47, Burning Bridges, Crazy Frog, dance duets and more. Ready-to-share MP4 in minutes, from $9.99.