A basic unit of data processed by a language model, representing words, word fragments, punctuation, or characters.
A token is a unit a language model uses to process information. In text, it may represent a whole word, part of a word, punctuation or other character sequences. Tokens are not the same as words, and the way text is divided depends on the model’s encoding and the language.
Token counts matter when evaluating input capacity, output limits and usage. A long document, instructions and conversation history can all contribute to the material being processed. The visible answer alone may not describe all usage in a particular system, so avoid estimating a bill or capacity from its word count without checking the relevant documentation.
For business content work, the practical issue is choosing the information the model actually needs. Supplying an entire archive when only one approved specification is relevant can make the task harder to manage. Reducing unnecessary material should not mean removing the evidence or caveats required for an accurate answer.
Takeaway: Treat tokens as model-specific processing units, not a universal word allowance. Check actual counting and limits when planning large-document or high-volume workflows.
Example
A team wants to summarize a collection of service manuals. Instead of assuming that a fixed number of pages will fit, it checks the relevant model’s limits and separates the task into coherent documents with clear source references. It preserves the details needed for each summary rather than trimming every manual arbitrarily.
What to check
Use a provider’s supported counting method where available. Keep input and output limits distinct, and check how the application handles excess material. Do not assume that a rough English-language estimate applies equally to code, tables or every other language.
For more detail, see OpenAI token documentation.