A clearer view of AI.
THE INDEPENDENT AI.

FRONTIER LAB / LLM · WORTH A LOOK

Liquid AI

Efficient Liquid Foundation Models (LFMs) for on-device, edge and cloud AI, plus d1 decision models.

Visit liquid.ai ↗@liquidai on X ↗

What Liquid AI does

Efficient Liquid Foundation Models (LFMs) for on-device, edge and cloud AI, plus d1 decision models. Open-weight models (LFM3 family launched 27 Oct 2026, LFM2.5, d1 decision models) that run locally on llama.cpp, MLX, ONNX, Ollama and LM Studio for private, low-latency AI, plus a paid API.

Good for

  • Private on-device assistants
  • Cheap fast decision/classification calls
  • Edge AI in products
  • Small offline model for local agents (LFM3-1.7B in Ollama, LM Studio or Osaurus)

Try this in 10 minutes

  1. Private on-device assistants. Use the free plan: pick one real example from your own business and try just that.
  2. Cheap fast decision/classification calls. Put the result next to how you do it today. Is it faster, cheaper or better?
  3. Edge AI in products. Note what you had to fix by hand before you'd show it to a customer.

Mini-tasks based on common small-business uses for this tool.

Honest take

Some claims not independently checked

  • Low-cost option for automations; available on OpenRouter and Vercel.
  • LFM3 (27 Oct 2026): LFM3-1.7B has 1.73B parameters, a 256K-token context, and needs about 1 GB of memory at 4-bit (Ollama download 1.1-1.7 GB).
  • Liquid measured a full 128K-token context at 1.6 GB of extra memory in llama.cpp.
  • The model card says to use retrieval or search for knowledge-heavy tasks and does not recommend it for agentic coding or complex multi-step reasoning.
  • Weights on Hugging Face (GGUF, MLX, ONNX), Ollama (ollama run lfm3) and LM Studio.
  • Licence: LFM Open License v1.0, based on Apache 2.0 but free commercial use only for organisations under $10M annual revenue (counting affiliated companies); research and non-commercial use is free.
  • On its own tables LFM3-1.7B leads on tool use and instruction following but trails Qwen3.5-4B on 128K long-context recall (RULER 65.6 vs 84.6).

What we couldn’t verify

  • Liquid's LFM3 benchmark claims are self-reported: the tables on the blog and model cards were run by Liquid, and the 'Pareto frontier of device intelligence' ranking is on pipette.liquid.ai, Liquid's own site.

Our Hype or Fact? checks

Pricing

API per-token (e.g. d1 $0.02/M input; LFM2.5-24B $0.05 in / $0.70 out per M); $2 free credit; open weights downloadable.

Prices as listed by the maker when we last checked (2026-10-29); usually in US dollars. Check the website for current prices.

Latest update

  • 2026-10-27: LFM3 launched: six open-weight ~2B models (LFM3-1.7B text with 256K context, LFM3-VL-2B vision, Audio, Omni, d1-2B and d1-Omni-2B decision models). About 1 GB of memory at 4-bit; free commercial use only under $10M annual revenue.

Is Liquid AI worth it? Quick answers

Is Liquid AI worth it?

Liquid AI is worth a look if it matches a job you already do (Frontier lab / LLM). Low-cost option for automations; available on OpenRouter and Vercel. There is a free tier, so you can test it before paying. Some of the maker's claims have not been independently checked.

How much does Liquid AI cost?

API per-token (e.g. d1 $0.02/M input; LFM2.5-24B $0.05 in / $0.70 out per M); $2 free credit; open weights downloadable.

What is Liquid AI used for?

Efficient Liquid Foundation Models (LFMs) for on-device, edge and cloud AI, plus d1 decision models. Typical uses: private on-device assistants; cheap fast decision/classification calls; edge AI in products.