Liquid AI

What it does

Liquid AI publishes open-weight foundation models (LFMs) in text, vision, audio, omni, and decision families, with context lengths documented per model, mostly 32K tokens. Developers provide text, images, audio, or video depending on the model, and receive text or speech output. Models are distributed as GGUF, MLX, and ONNX packages for local and device runtimes, and can also be called through an API.

People typically choose a model family matching their input and output shape, then a runtime such as llama.cpp, Ollama, MLX, ONNX Runtime, vLLM, or SGLang. Fine-tuning is supported through Liquid Finetune, TRL, or Unsloth. The distinct focus is small footprint and efficiency for on-device work, with documented quantization options such as Q4_K_M.

This suits developers and enterprises building latency- or privacy-sensitive products that run on phones, laptops, cars, or edge hardware. Open-weight models are free to use, including commercially, for organizations under $10M annual revenue, with a paid enterprise license above that threshold. API usage is billed per million tokens, and each model page lists exact limits and supported runtimes.

What it can do

• Text chat, tool calling, structured output, and classification with text-to-text models
• Image understanding with vision-language models
• Interleaved audio/text models for speech and voice chat
• Decision models for classification, routing, and scoring
• Fine-tuning open-weight models with Liquid Finetune, TRL, or Unsloth
• Deployment as GGUF, MLX, or ONNX packages for local and device runtimes

Works with

• llama.cpp
• MLX
• ONNX Runtime
• Ollama
• SGLang
• vLLM
• OpenRouter
• Vercel AI Gateway

Before you choose

• Context length varies by model (for example 256K for LFM3-1.7B, 32K for most others); check each model page.
• Runtime support differs by model; audio, omni, and decision models support fewer runtimes.
• Open-weight free use applies only to organizations under $10M annual revenue; larger organizations need an enterprise license.

Listed under

• Coding
• Sandboxes and infrastructure
• Developers
• Edge AI
• Enterprise
• LLM API
• On-device AI
• Open weight models

Get started

Every new organization gets $2 in free credit on Liquid Console.

1. Choose a model family matching your input and output shape from the docs.
2. Select a runtime (llama.cpp, Ollama, MLX, ONNX, vLLM, or SGLang) based on your deployment target.
3. Download the matching GGUF, MLX, or ONNX repository from Hugging Face, or sign up on Liquid Console for API access.

API prices per million tokens (input, cached input, output) vary by model; check the pricing page rate table. Every new organization gets $2 in free credit on Liquid Console. Enterprise license pricing is custom above $10M annual revenue.

Source: liquid.ai/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Liquid AI | Device-native foundation models. ↗www.liquid.ai
• Pricing | Liquid AI ↗www.liquid.ai
• Liquid Foundation Models - Liquid Docs ↗docs.liquid.ai

Company website

liquid.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.