Back to glossary

Token

Token

D2

In OpenAI’s documented text processing, a token is a unit into which text is divided for a model to process or generate.

Review status: 2026-11-26

Technical explanation

In OpenAI’s documentation, a token can represent a character, part of a word, a whole word, punctuation, or spacing depending on encoding and context.

Conceptual boundaries

In OpenAI’s documentation, a token is not necessarily a word, and token counts can differ for the same text across encodings and languages.

Provider-neutral example

Before submitting a long request, a team can count the tokens produced by the target model’s tokenizer instead of estimating only its word count.

Limitations

Token examples and counts are encoding-specific, so a count from one model does not automatically apply to another model.

Related concepts

Atomic claims and evidence

  1. 1.1OpenAI Help Center, Understanding and counting tokens
    Source
    OpenAI Help Center, Understanding and counting tokens
    Source role
    Authoritative source
    Exact locator
    Overview: token units and possible text segments
    Supported claim
    Tokens are units that a model uses to process text, and a token can represent a character, part of a word, a whole word, or punctuation.
    Last verification
    Review due
    Scope limitation
    This is official technical documentation for its models’ text processing and does not prescribe one tokenizer for all language models.
  2. 2.1OpenAI Help Center, Understanding and counting tokens
    Source
    OpenAI Help Center, Understanding and counting tokens
    Source role
    Authoritative source
    Exact locator
    Overview and Understand how text becomes tokens: count variability
    Supported claim
    For OpenAI’s documented text models, token count is not the same as word count and can vary with the model, encoding, and language.
    Last verification
    Review due
    Scope limitation
    This is provider-bounded documentation and does not establish a fixed conversion ratio or universal tokenizer behavior.