LFM2.5-24B-A2B: 24B parameters, 2B active for on-device agents
LiquidAI released LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters that produces chains of thought, calls tools, and is designed to run agents on-device.
What changed
LiquidAI released LFM2.5-24B-A2B: 24B parameters, 2B active parameters, 40 layers, 64 experts, 131K context window, 128K vocabulary; reasoning-only model that outputs an explicit chain of thought.
Why it matters
Developers can run private, continuous agents on local hardware because only 2B parameters are active per token, lowering per-token inference cost for tool-using, multi-step workflows.
Who should care
DevelopersEnterprises
Worth trying
Yes — Open-weight on Hugging Face; runs on M5 Max and B300 GPU.