The basic unit of text that AI models process — roughly equivalent to a word or word fragment.
Tokens are how AI models read and write text. A tokenizer splits text into tokens before the model processes them, and the model generates tokens one at a time when producing output. One token is roughly 3-4 English characters, so "artificial intelligence" is about 4-5 tokens.
Tokens matter for three practical reasons. First, they determine cost — API pricing is per token (input and output). Second, they define the context window limit — how much text the model can process at once. Third, they affect speed — generating more tokens takes more time.
Understanding token economics is essential for building cost-effective AI systems. A poorly designed agent that stuffs unnecessary context into every request wastes tokens (and money). A well-designed agent retrieves only what it needs, structures prompts efficiently, and caches results. At Agentik {OS}, token efficiency is a design principle — our orchestration layer optimizes what information each agent receives, keeping costs predictable while maintaining quality.
Want to see AI agents in action?
Book a Demo