What's this about?
South Korean telecom SKT released A.X K2, a new open-weight large language model, on Hugging Face — a large sparse mixture-of-experts model with 688 billion total parameters but only 33 billion active at any time.
What exactly is a "sparse MoE" with "active" parameters?
A mixture-of-experts (MoE) model is split into many specialist sub-networks ("experts"). For any given input, a router picks only a few experts to actually run — so a 688B model might only use 33B of them per token. "Sparse" means most of the model sits idle each step. The payoff: you get the knowledge capacity of a huge model while paying the compute cost of a much smaller one.
Why is this worth noting?
It's another entry in a fast-widening field of open-weight frontier-class models coming out of Asia (Kimi, GLM, and now SKT), reinforcing the week's dominant theme that the open-weight frontier is no longer a one-lab story.