tools
Liquid Foundation Models
Open foundation models for text, vision, audio, and task-specific AI that run locally on phones, laptops, edge devices, and servers.
Liquid Foundation Models are open-weight models for text, vision-language, audio, embeddings, extraction, translation, RAG, and other specialized tasks. They are designed for inference on CPUs, GPUs, and NPUs across local, cloud, hybrid, and on-device deployments.
Developers can download, run, fine-tune, and commercially deploy the models free under the LFM Open License while their company has less than $10 million in annual revenue. Larger companies need a commercial license; custom models, enterprise support, and deployment optimization are handled through Liquid AI.
Features
- Text models for chat, tool calling, structured output, reasoning, and classification
- Vision-language models for image understanding and multimodal tasks
- Audio-language models for speech, audio, and text generation
- Nano models for extraction, summarization, RAG, translation, and embeddings
- Runs on CPUs, GPUs, and NPUs across phones, laptops, vehicles, and edge devices
- Supports deployment with llama.cpp, MLX, ONNX, CoreML, SGLang, and vLLM
- Free commercial use under $10M annual company revenue
- LFM Open License permits modification, distribution, and private fine-tunes
Use cases
- Build private assistants that run locally on phones and laptops
- Deploy low-latency agents for vehicles, robots, and embedded devices
- Fine-tune models for proprietary company data and domain workflows
- Extract structured data from documents, images, and audio
- Run local RAG, translation, summarization, and classification systems
- Add speech and multimodal interaction to edge applications
Pros
Cons
Latest updates
- Introducing LFM3: Every Modality for Specialized Intelligence (LFM3)
Every Modality for Specialized Intelligence
- d1-ultra: Introducing the World’s Best Decision Model
Introduces d1-ultra, a decision model.
- LFM2.5-24B-A2B: Built for Fast On-Device Agents
Built for on-device agents.
- Open d1: Edge decision models for text, vision, and audio
d1 for text, vision, and audio.
- Introducing d1: The most capable decision model, now with vision
Capabilities
- Choice of models — “Access our models via API, license our open weight models to build on your own, or partner with us on a custom deployment” source
- Self-hosted — “License our open weight models and run them on your own infrastructure” source
- API — “The fastest way to start. Get a key from Liquid Console, then call our models over an OpenAI-compatible endpoint.” source
- Runs models for you — “The fastest way to start. Get a key from Liquid Console, then call our models over an OpenAI-compatible endpoint.” source
Get it
Pricing
- Starting price
- Free
- Prices checked
- 2026-10-24
d1
- Input $0.02 / MTok
- Cached input $0.02 / MTok
- Output $0.00 / MTok
d1-ultra
- Input $0.042 / MTok
- Cached input $0.042 / MTok
- Output $0.00 / MTok
LFM2.5-24B-A2B
- Input $0.05 / MTok
- Cached input $0.025 / MTok
- Output $0.70 / MTok
Free
Free
- Use open weight models in production at no cost for businesses with less than $10M in annual revenue
- Download, run, and fine-tune every open LFM
- Deploy anywhere: mobile, edge, on-prem
- Keep your fine-tunes private; no copyleft
- Research, education, and non-profit use: always free, with no revenue limit
Enterprise license
Price on request
- For businesses with more than $10M in annual revenue
- Download, run, and fine-tune every open LFM
- Deploy anywhere: mobile, edge, on-prem
- Keep your fine-tunes private; no copyleft
- Pricing is structured around your deployment scale
Custom Deployment
Price on request
- Bespoke architecture and optimization
- OEM and on-prem deployment support
- Dedicated support and SLAs
- Your data never leaves your environment, and we never use customer data to train models
- Pricing is structured around your deployment scale and support needs