AIToday
Large Language ModelsOpen-Source AIGIGAZINE AIPublished: Oct 14, 2026, 19:01 JST

Liquid AI launches LFM2.5-24B-A2B for on-device agents

Liquid AI launches LFM2.5-24B-A2B for on-device agents

3 Key Points

  1. What happened

    Liquid AI released LFM2.5-24B-A2B, a 24-billion-parameter MoE model with 2.45 billion active parameters, pretrained on 33 trillion tokens and tuned for agent workflows.

  2. Why it matters

    The model scores higher than larger rivals like Gemma-4-26B-A4B-it and Qwen3.6-35B-A3B on agent benchmarks, and is pitched as an ideal base for on-device agents.

WHO IT HITSDevelopers and IT teams building local AI agents on laptops or desktops can run LFM2.5-24B-A2B without cloud API calls, as the demo shows Hermes Agent running fully on-device.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

Liquid AI's LFM2.5-24B-A2B is a 24-billion-parameter Mixture-of-Experts model with only 2.45 billion active parameters, pretrained on 33 trillion tokens and further trained for agent workflows. It is positioned against larger models like Qwen3.6-35B-A3B and Gemma-4-26B-A4B-it, and it scores higher on agentic tests such as AutomationBench and BFCLv4. Because llama.cpp, MLX, vLLM, SGLang, and ONNX already support it, the model can run locally on Windows and Mac. A demo shows Hermes Agent running LFM2.5-24B-A2B on a single laptop with no cloud API calls or keys.

FAQ
What is the model size of LFM2.5-24B-A2B?
LFM2.5-24B-A2B has 24 billion total parameters and 2.45 billion active parameters in a MoE architecture.
Which tools support LFM2.5-24B-A2B inference?
llama.cpp, MLX, vLLM, SGLang, and ONNX already support it, enabling use on Windows and Mac.
How fast does LFM2.5-24B-A2B run?
On a MacBook Pro with M5 Max, the Q4_K_M quantized version decodes at 184 tokens per second using 15GB of memory.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMekom unveils DX diagnosis service from new Yamagata HQ