Token

A Token is the small chunk of text that a language model processes. Models do not see whole words or sentences; they break input into tokens, which are often parts of words, and generate output one token at a time as they assemble a response. Tokens drive both the cost and the capacity of AI systems, which makes them a budget item, not just a technical detail.

What Token Means

A Token is the basic unit of text a language model reads and generates, usually a word fragment, which also determines how usage is measured and priced. A rough guide is that one Token equals about four characters of English text, so 750 words is roughly 1,000 tokens. Both the prompt you send and the response you receive count toward usage, and longer outputs cost more even when the input is identical. Common words may be a single Token, while rare words, names, or technical terms get split into several. Languages other than English and unusual formatting also increase token counts, sometimes substantially, which catches teams off guard when localizing or processing dense technical content.

How a Token Works

A language model uses a tokenizer to convert text into Tokens before processing. Different models use different tokenization schemes, so the same passage may use more tokens in one model than another. At inference time, the model reads input tokens, processes them through its layers, and generates output one token at a time, with each new token influenced by everything that came before. The context window is the maximum number of tokens a model can consider in a single request, including the prompt, attached documents, conversation history, and the response itself. Most pricing separates input tokens from output tokens, and output is often more expensive, which shapes the economics of high-volume content generation.

Common Pitfalls and Misconceptions

A common misconception about Tokens is that token limits refer to word counts. They do not, and dense or technical text can consume more tokens than expected. Another pitfall is estimating cost based on input length alone when output tokens are usually priced higher and can dominate spend on generation-heavy workflows. A third is using the most capable model for tasks that a cheaper one would handle adequately, which is one of the most common sources of avoidable AI spend. Rerouting tasks by complexity can cut costs noticeably without affecting quality, and reviewing token usage by workflow surfaces where the spend actually concentrates.

Token in Practice

The practitioner habit that controls AI spend in mature programs is treating Tokens like any other variable cost. Teams instrument token usage by workflow, watch the ratio of input to output tokens, and notice when prompts grow longer than necessary or when models start producing more verbose responses than the task requires. A few percent reduction in tokens per task compounds dramatically across millions of operations, which is where AI cost discipline either lives or quietly slips beyond what was budgeted. Trimming system prompts to essentials, summarizing long source material, constraining output length, and avoiding unnecessary back-and-forth are the optimization habits that keep costs predictable.