Alibaba Just Dropped Marco-Mini and Marco-Nano: Sparse AI Models That Run Stupidly Fast

Alibaba quietly released two new AI models six days ago that nobody's really talking about yet. Marco-Mini has 17.3 billion parameters but only activates 0.86 billion at a time, and Marco-Nano is even smaller at 8 billion with 0.6 billion active. They use a Mixture-of-Experts (MoE) architecture, which basically means the model only "wakes up" the parts it needs for each task—like having a team of specialists instead of one person doing everything. The parameter-to-active ratio is genuinely wild, which means these should be lightning fast and cheap to run. Perfect if you want decent multilingual performance without needing a data center.