Liquid AI
Compute-efficient, downloadable foundation models for on-device, edge, private-infrastructure, and cloud deployments.
About Liquid AI
Liquid AI builds compute-efficient foundation models for mobile devices, edge hardware, private infrastructure, and cloud servers. Its open LFM families cover text and vision-language workloads and can be downloaded, run, and fine-tuned. Companies with less than $10 million in annual revenue may use them commercially at no cost under the LFM Open License. Above that threshold, an enterprise license is required; pricing depends on deployment scale and support needs. The license has no copyleft requirement, so fine-tunes can remain private. Ollama supports straightforward local serving, while SGLang provides low-latency, high-concurrency serving through an OpenAI-compatible API on CUDA GPUs. Liquid AI suits teams that care about model footprint, local control, or edge execution. Hardware requirements and runtime compatibility still need careful review before a serving stack is selected.
Best for
Best use cases
Run compact AI models on mobile, automotive, or edge hardware
Fine-tune a model for private product cataloging or domain workflows
Serve LFM2.5 at high concurrency through an OpenAI-compatible API
Compare before choosing
The free commercial threshold reduces early-stage licensing cost
Self-hosting provides more deployment control than a vendor-only hosted API
Runtime compatibility and hardware requirements require more engineering review than a managed API
Features
On-device AI
Run compact text or vision-language models on phones, laptops, vehicles, and other constrained devices.
Private model customization
Fine-tune an open LFM for a specialized workflow while keeping the resulting fine-tune private.
Production model serving
Serve LFM2.5 through an OpenAI-compatible API with SGLang on CUDA GPUs or use Ollama for simpler local deployment.
Pros
- Commercial model use costs $0 while company annual revenue remains below $10 million.
- Models can run on mobile, edge, on-premises, or cloud infrastructure, giving teams deployment control.
- The LFM Open License has no copyleft requirement, so private fine-tunes can remain private.
Cons
- Companies above $10 million in annual revenue must negotiate a custom enterprise license.
- Liquid AI does not provide its own hosted API, so teams may need self-hosting or a third-party service.
- Some serving paths have specific constraints: SGLang requires a CUDA GPU, and LFM MoE models need a compatible Ollama version.
Pricing Plans
Free
- Download, run, and fine-tune every open LFM
- Commercial use for companies under $10M in annual revenue
- Deploy on mobile, edge, or on-premises infrastructure
- Keep fine-tunes private with no copyleft requirement
Enterprise
- Commercial license beyond the $10M annual revenue threshold
- Bespoke architecture and optimization
- OEM and on-premises deployment support
- Dedicated support and SLAs
* Pricing information is for reference only. Please refer to the latest announcement on Liquid AI's official website.