Mixture of Experts (MoE)
An AI architecture where multiple specialized sub-networks (experts) handle different parts of a task. A gating network routes each input to the most relevant experts, achieving high performance with fewer active parameters per query.
Why it matters
Mixture of Experts lets a model be huge yet fast, because only the relevant sub-networks activate for any given input rather than the entire model. This is how some of the newest frontier models deliver strong performance without a proportional jump in cost per response. Recognizing the term helps you understand why certain large models feel surprisingly efficient and why "total size" alone doesn't tell the whole story.
A concrete example
Think of a hospital where a receptionist routes each patient to the right specialist instead of every doctor examining everyone. In an MoE model, a gating network sends a coding question to the parts trained on code and a poetry request elsewhere. You never see this happen, but it's why a model with hundreds of billions of parameters can still reply quickly and at reasonable cost.
How to use it
The practical consequence is a model with a very large total parameter count that costs far less to run than that number suggests, because only part of it activates per token. When comparing models, this is why total size is a poor guide to speed or price, and why you should compare on what a model actually costs and how it performs on your task rather than on headline parameter counts. Switch Transformers is the clearest write-up of why this scales the way it does.
The common mistake
Comparing a sparse model's total parameters against a dense model's as though they mean the same thing. They do not, and the comparison usually flatters whichever number the marketing chose.
Related terms
Inference Cost
The computational expense of running a trained AI model to generate outputs. Measured in cost per token or per request. Varies dramatically between models — GPT-4 class models cost 10-50x more per token than smaller models.
Model Routing
Automatically selecting the best AI model for each query based on task type, complexity, cost, and latency requirements. Smart routers analyze your prompt and route it to the optimal model — saving money on simple tasks and ensuring quality on complex ones.
Transformer
The neural network architecture behind modern AI models like GPT and BERT. Uses self-attention mechanisms to process entire sequences in parallel, enabling much faster training than previous architectures. Nearly all large language models are transformer-based.
AI Dubbing
Automated translation and re-voicing of audio/video content into other languages while preserving the original speaker's voice characteristics, timing, and emotional delivery.
AI Orchestration
Coordinating multiple AI models, tools, and data sources in a unified pipeline. An orchestration layer manages prompt routing, context passing, error handling, and output aggregation across different AI services.
API (Application Programming Interface)
A way for software applications to communicate with each other. AI APIs let developers integrate AI capabilities into their own applications programmatically.