LFM2-8B-A1B just dropped on @huggingface!

8.3B params with only 1.5B active/token 🚀

> Quality ≈ 3–4B dense, yet faster than Qwen3-1.7B
> MoE designed to run on phones/laptops (llama.cpp / vLLM)
> Pre-trained on 12T tokens → strong math/code/IF pic.twitter.com/cGbuJoOMDN

— Maxime Labonne @ ICLR (@maximelabonne) 7 octobre 2025

LFM2-8B-A1B just dropped on @huggingface
! 8.3B params with only 1.5B active/token > Quality ≈ 3–4B dense, yet faster than Qwen3-1.7B
> MoE designed to run on phones/laptops (llama.cpp / vLLM)
> Pre-trained on 12T tokens → strong math/code/IF

→ View original post on X — @maximelabonne,