ByNisha Talagala,
Contributor.
As AI revolutionizes all industries, businesses have had to grapple with a new term and concept, tokens. Tokens are now a foundational expense for businesses as well as a critical item in return on investment and employee performance assessment. Terms like tokenmaxxing, the business behavior of maximizing token assets, are spreading beyond Silicon Valley. Companies are publicly assessing employees by token use, with both too high and too low values generating concern. What are tokens, and what should your business leaders know about them?
A token is the unit of an AI’s processing. Originating from the natural language processing technology at the heart of ChatGPT and other large language models, a token is either a group of words, a word, or part of a word. Each AI is capable of processing some number of tokens at a time (called the content window). The AI services are billed by tokens, and depending on the level of service purchased, may have fixed or variable token prices.
Say that an employee is using an LLM to revise an email. The first message, including the draft email test and instructions, will contain a number of tokens. The response will as well. Any follow-on messages will be added to the context window and sent back as additional tokens. Studies have shown that a simple task like this consumes around 100 tokens per email on common LLMs like ChatGPT, Claude, or Gemini. At the time of this writing, common token prices would suggest this task costs around 0.25 US cents.
