Mixture of Experts (MoE)
A model design that routes each request to only the relevant parts of the network instead of using the whole thing every time.
It lets a model be huge and knowledgeable while staying fast and cheaper to run.
Frequently asked questions
What does mixture of experts mean without the jargon?
It is a model design that only wakes up the relevant parts of its brain for each request instead of using the whole thing every time. Like a big firm that routes your question to the one specialist who handles it, rather than the entire staff.
Why does mixture of experts matter to me as a user?
It is a big reason some very large, knowledgeable models still respond quickly and cost less to run. By activating only the needed parts, the model can be huge without being slow or expensive on every single answer.
Does a mixture of experts model mean there are real separate experts inside?
Not in the human sense. The experts are different sections of one network, and the model automatically picks which sections to use for a given input. There is no team of people, just a smarter way of dividing up the work internally.
New to all this? Start with what an AI agent really is, browse the full glossary, or explore the learning hub.
← Back to the glossary