Liquid AI
Developers, researchers, startups, and enterprises that need efficient, on-device AI for applications in automotive, consumer electronics, e-commerce, financial services, healthcare, industrial, and defense sectors.
Overview
Liquid AI is an efficiency-first foundation model company spun out of MIT CSAIL. Founded in 2023 by Ramin Hasani (CEO), Mathias Lechner (CTO), Alexander Amini (CSO), and Daniela Rus, the company builds general-purpose AI systems that are compute-optimized and designed to run on devices outside of data centers. Liquid AI has raised a total of $297 million in funding over two rounds, with its latest round being a Series A. The company is headquartered in Cambridge, MA.
What it does
Liquid AI develops Liquid Foundation Models (LFMs), a family of device-native foundation models that run on phones, laptops, cars, and embedded devices. The models are designed for on-device inference, keeping data on the device and reducing latency. LFMs range from a few hundred million to a few billion parameters, with variants small enough to run under 1GB. The company also offers the Liquid Edge AI Platform (LEAP) SDK for fine-tuning and deploying models, and runs Liquid Labs for frontier research in liquid neural networks and state-space models.
Key features
- Device-native foundation models (LFMs) that run on CPUs, GPUs, and NPUs
- On-device reasoning under 1GB
- Fine-tuning and deployment via LEAP SDK
- Support for multiple runtimes: llama.cpp, MLX, ONNX, CoreML, SGLang, vLLM
- Open license with no copyleft, allowing private fine-tunes
- Free commercial use under $10M annual revenue
- Model variants from 230M to 24B parameters, including Mixture of Experts (MoE)
- Multimodal capabilities (LFM2.5-VL-450M for visual intelligence)
Use cases
- On-device AI for consumer electronics and mobile apps
- Embedded in-car intelligence for automotive
- Drug discovery with lightweight scientific foundation models
- Edge AI for e-commerce, financial services, and industrial applications
- Privacy-sensitive workloads where data must not leave the device
- Tool-calling agents on consumer hardware
Pricing
| Plan | Price |
|---|---|
| Free | Free |
| Enterprise | — |
Models are free to download, run, and fine-tune, including commercially, until company's annual revenue passes $10 million USD. Above that, a commercial license is required. Research, education, and non-profit use are always free with no revenue limit.
View current pricingPricing is gathered from the company's public pages and may change. Check the vendor's site before buying.
Pros and cons
Strengths
- Free to download, run, and fine-tune for commercial use under $10M annual revenue
- No copyleft, so fine-tuned models can remain proprietary
- Runs on a wide range of hardware, including CPUs and edge devices
- Strong focus on efficiency and compute optimization
- Partnerships with major companies like Mercedes-Benz and Insilico Medicine
- Backed by significant funding ($297M) and spun out of MIT CSAIL
Limitations
- Pricing for enterprise tier is not publicly disclosed; requires contacting sales
- Free commercial use limited to companies under $10M annual revenue
- No self-serve enterprise option; enterprise support requires sales engagement
- Model sizes may be limited compared to larger cloud-based models
What sets it apart
- Efficiency-first approach with models designed to run on-device under 1GB
- Open license with no copyleft, allowing private fine-tunes
- Strong focus on edge and embedded deployment, not just cloud
- Backed by MIT CSAIL research in liquid neural networks and state-space models
Ecosystem
Integrations
Notable customers
Frequently asked questions
Is Liquid AI really free to use?
Yes. Liquid's open foundation models are free to download, run, and fine-tune, including in commercial products, until your company passes $10 million in annual revenue. The full terms are in the LFM Open License.
Can I use Liquid AI in a commercial product?
Yes. The license permits commercial use; the only limit is the $10 million annual revenue threshold. Above it, you'll need a commercial license.
If I fine-tune a model, do I have to release it?
No. The license has no copyleft requirement, so you can keep your fine-tuned models private and proprietary.
What about research, universities, and non-profits?
Qualified non-profits, schools, and researchers can use the models for non-commercial and research purposes for free, with no revenue threshold.
What happens when we pass $10 million in revenue?
Your free commercial rights end and you'll need a commercial license to keep running the models in production. You can contact Liquid AI to find the right path.
Will Liquid models run on my hardware?
LFMs are built to run on CPUs, GPUs, and NPUs, from phones and laptops to vehicles and embedded devices. Model sizes range from a few hundred million to a few billion parameters, with variants small enough to run under 1GB.
Sources
This profile was compiled from the company's own pages and public web research.
Last researched August 11, 2026.
Manage this listing
Own this company? Update your listing information.