Liquid AI
Liquid AI builds device‑native foundation models (LFMs) that run locally across phones, laptops, vehicles, and embedded systems. A compact model family, exportable to llama.cpp, MLX, ONNX, and CoreML, enables private, low‑latency reasoning and search with 3200+ variants and industrial deployments.
Overview
Pick an LFM variant sized to your target hardware, fine‑tune with LEAP on domain data, quantize, then export a single artifact per runtime. Deploy to laptops, mobiles, or vehicles with private, on‑device inference and monitoring through your existing application stack.
Platform overview
Built for engineers shipping AI into products beyond the datacenter: mobile app developers, automotive software teams, embedded and robotics engineers, and ML practitioners in e‑commerce, finance, healthcare, industrial, and defense. Researchers exploring liquid neural networks, state‑space models, and tokenizer upgrades can prototype quickly, then validate with Pipette benchmarking. Organizations in regulated environments adopt LFMs to keep sensitive content local while matching interactive responsiveness users expect on consumer hardware.
- Fine-tune LFMs with LEAP, then export to MLX, ONNX, or CoreML.
- Deploy sub‑1GB models to laptops, mobiles, and vehicles for private inference.
- Use LFM2.5‑VL for multimodal understanding, grounding text with images on‑device.
- Accelerate inference with DSpark and quantization‑aware distillation for edge deployment.
- Run retrieval and long‑context encoders efficiently, even on CPU‑only systems.
Highlights
Who is it for
Start by exploring the Liquid Foundation Models catalog and Model Playground to benchmark candidates. Install the LEAP SDK, load a base LFM, and fine‑tune on curated datasets. Use provided recipes for quantization‑aware distillation, tokenizer expansion, and retrieval alignment. Bake runtime artifacts and export to llama.cpp, MLX, ONNX, or CoreML for your target hardware; SGLang and vLLM cover server or hybrid needs. Validate with Pipette, then push to production on laptops, phones, or vehicles. Documentation and example projects shorten bring‑up on Apple Silicon and CPU‑only endpoints.
Bring frontier models to local processors—no data center required.
Getting started
Liquid AI differentiates through breadth of shipped LFMs, on‑device performance under 1GB, and a practical path from research to deployment. The combination of LEAP, exportable runtimes, and production partnerships shows maturity. If you need private, low‑latency intelligence across heterogeneous edge hardware, with options from encoders to multimodal MoE, Liquid AI is purpose‑built.
Open the tool and review its core product experience.
Create your account or access your existing workspace.
Use your own task to judge speed, quality, and fit.
Check similar AI tools before making a final decision.
Liquid AI
Liquid AI is a research‑driven company founded in 2023 to deliver efficient, general‑purpose intelligence on edge hardware. The team ships Liquid Foundation Models, the LEAP SDK, and open research that advances quantization, distillation, and structured outputs for real‑world deployments across industries.
Liquid AI provides a family of device‑native foundation models (LFMs) designed for edge hardware where latency, privacy, and power budgets matter. Models span compact encoders and general‑purpose language models to multimodal LFM2.5‑VL, and mixture‑of‑experts variants, many small enough to run under 1GB. The toolchain centers on the LEAP SDK to fine‑tune on your data, then export baked artifacts to popular edge runtimes including llama.cpp, MLX, ONNX, CoreML, SGLang, and vLLM. Production deployments already cover phones, laptops, vehicles, retail, and industrial settings. Recent research releases include DSpark performance improvements and Q4_0 quantization‑aware distillation for efficient inference, alongside retrievers and encoders optimized for long context on CPU. With 56 LFMs shipped, 41.7M downloads, and 3200+ variants, Liquid AI focuses on bringing frontier‑grade intelligence to real‑world hardware. Partnerships with automotive and silicon leaders underscore production readiness.
Visit Liquid AI
Comments (0)
No Comments Found