Liquid AI releases LFM2.5-24B-A2B open-weight model for on-device agents

Shows a small active-parameter MoE can run capable agent workloads privately on-device.

LFM2.5-24B-A2B is an open-weight Mixture-of-Experts model from Liquid AI with 24B total and about 2B active parameters, designed to run fast, private agent tasks locally on a laptop. Weights are available on Hugging Face.

Name

LFM2.5-24B-A2B

Params

24B (A2B active)

Read it at Hugging Face Blog →

Coverage: 8 sources, in order

1. 13 Oct 17:05 Hugging Face Blog: Run fast, private agents on-device with LFM2.5-24B-A2B
2. 13 Oct 17:14 Reddit: LiquidAI/LFM2.5-24B-A2B
3. 13 Oct 17:29 Unite.AI: Liquid AI Releases Open-Weight LFM2.5-24B-A2B for On-Device Agents
4. 14 Oct 01:20 Startup Fortune: Liquid AI's LFM2.5-24B-A2B Runs Agent Tasks Locally on a Laptop
5. 14 Oct 06:37 @DogukanUrker on X: ran my agentic coding benchmark on LFM2.5-24B-A2B. same 20 PatchEval tasks, Pi agent, single RTX 3060. 8/20. Qwen3.8-27B and Qwen3.6-35B-A3B both get 16, Ornith-1.5-35B gets 15. Coding isn’t where it shines vs tool use.
6. 14 Oct 07:02 Hugging Face: LiquidAI/LFM2.5-24B-A2B (model)
7. 14 Oct 07:50 MarkTechPost: Liquid AI Releases LFM2.5-24B-A2B: A 2B-Active MoE for On-Device Agents
8. 14 Oct 15:00 Hugging Face: LiquidAI/LFM2.5-24B-A2B-GGUF (model)

← Front page