Artificial intelligence token efficiency relates useful task outcomes to the amount of tokenized text processed or generated. A system can improve efficiency by using focused context, concise outputs, suitable models, caching, retrieval, or workflow designs that avoid repeated and unnecessary model calls.
Token efficiency affects more than billing. It can influence latency, throughput, context-window usage, and infrastructure capacity, although it should be assessed alongside quality because the fewest tokens do not automatically produce the best result.
