Liquid AI LFM2.5-24B-A2B in chat: a 2.45B-active reasoning MoE
Liquid AI's LFM2.5-24B-A2B is available in FastMetal chat. It is the largest model in the LFM2.5 series and also the last one, and Liquid AI built it for agents that run fast on local devices.
LFM2.5-24B-A2B is a chat-only model on FastMetal: you use it in the web chat, and it is not available through the API or MCP.
What it is built for
- Speed from a small active set. The model has 24B parameters in total, but each token uses only 2.45B of them, so it is light and quick to answer.
- Reasoning on every request. It always thinks before it answers, and the thinking counts as output tokens.
- Tool-using agent work. Liquid AI reports a LiveClaw score of 48.09, ahead of Qwen3.6-35B-A3B at 38.19, a model about 1.5 times its size in both total and active parameters. These are Liquid AI's own figures.
- Work on material you supply. Liquid AI says the model is weak on knowledge-heavy questions alone and recommends pairing it with search. Paste in your documents and ask it to organize them, or have it plan the steps of a task.
- A 128K context. We serve it with 128,000 tokens, and Japanese is among its supported languages.
Pricing
LFM2.5-24B-A2B is billed per million input tokens and per million output tokens. Current rates are on the model page.
Billing is prepaid: you load a balance by card, from $5, and you can never spend more than you loaded. See pricing.
How to use it
Compare real answers first
The model page shows LFM2.5-24B-A2B's actual answers to our shared prompts: summaries, bug fixes, translation and more. Compare them with the model you use today before you switch. Everything we serve is in the model catalog.