Vision expert Language expert Code expert Math expert Audio expert Logic expert
⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.

Mixture-of-Experts Router: How One Model Handles Many Domains

Large "universal" AI systems don't run every parameter on every input — they route. This simulator renders a real gating network in 3D: pick an input domain, and watch a router score six specialist sub-networks, select only the top-k highest-scoring ones, and blend their outputs by renormalized softmax weight. A temperature slider reshapes how peaked or spread out that routing is, live gate-weight bars and Shannon entropy show exactly how confident the router is, and a running compute-cost readout demonstrates the core trick behind sparse Mixture-of-Experts models like Mixtral and Switch Transformer: a system with a huge total parameter count that only activates a small, efficient slice of itself per token.