Back to Glossary Index
Core ConceptModel Encoding Layer

Token

Industry Definition Set • Entity Resolution Path: /glossary/token

Quick Answer / TL;DR

The fundamental unit of text processed by LLMs, representing a word, subword, or character chunk that the model encodes and decodes.

Key Takeaways

  • Fundamental unit of text for LLMs.
  • Context windows are measured in tokens.
  • Billing is typically per-token for API usage.
  • A rough rule of thumb: 1 token ~ 4 characters in English.
Definitive Statement: The fundamental unit of text processed by LLMs, representing a word, subword, or character chunk that the model encodes and decodes.

Technical Context & Protocol Usage

Detailed Explanation
Tokens are the building blocks of LLM input and output. A token can be as short as a character or as long as a word, depending on the tokenizer. Models have a context window measured in tokens (e.g., 8K, 128K tokens). When a user sends a prompt, it is tokenized into a sequence of token IDs that the model processes. Billing for most LLM APIs is per-token.

Format & Payload Metadata

Format: Integer IDs mapped to vocabulary entries

Latency: Tokenization is near-instant; model inference scales with token count

Real-World Implementation Use Case

A 1,000-word prompt might be ~1,300 tokens, costing a fraction of a cent at current API rates.

M
MCPserver.in Engineering

Platform Team

Published: 2026-07-20
Updated: 2026-07-20

Cite This Page

MLA Style:

MCPserver.in Engineering. "Token." MCPserver.in Knowledge Hub, 20 July 2026, mcpserver.in/glossary/token.