Skip to content

Token

In the context of large language models, a token is the basic unit of text the model reads and generates. Tokens are not words—they are chunks of text determined by the model's tokenizer, typically three to four characters. The word "embedding" is one token. The word "unbelievable" is three. A period is one. This matters because everything in LLM-land is priced, measured, and constrained by tokens: API costs are per-token, context windows are measured in tokens, and rate limits cap tokens per minute. When someone says a model has a 128,000-token context window, that is the total budget for input and output combined. Understanding tokenization is not academic—it directly affects what you can fit in a prompt, how much a query costs, and why the model sometimes splits words in unexpected places.

Referenced in these posts:

  • Transformers are Eating the World

    Reflecting on three years of AI adoption, this talk emphasizes that transformative technologies like transformers are still in their infancy, requiring...

  • Things I Think I Think About AI

    Noah distills his 2,400+ hours of AI use into a candid, unordered list of 29 controversial takeaways—from championing ChatGPT’s advanced models and token...

  • Don’t Let SaaS Train on Your Private Tokens

    Don't let SaaS solutions train on your unique competitive advantage and protect your company's unique IP by building your own custom AI.

  • The $1 Sweet Spot

    Gemini Flash delivers around 80% of frontier LLM intelligence at just 10–25% of the cost, making it the defining model in the $1 intelligence tier.

  • The Cathedral, the Bazaar & the AI Economy

    AI's future demands both cathedral-style labs to push frontiers and bazaar-style open models to spread advances, boosting innovation and growth.

  • An Almanac for the Age of Chaos

    Marketing’s three-body problem: how to navigate when customers, channels and craft are all changing at once

  • Things I Think I Think About AI (2026 Edition)

    In the 2026 edition of Things I Think I Think About AI, I revisit AI harnesses, top models like Fable, and agent-first workflows reshaping the field.

  • On AI and Writing

    AI shouldn’t replace writing’s role as externalized thinking: true writing emerges from your unique voice, not a flattened, consensus-driven model.

  • The Three Vectors of the AI Race

    The AI race unfolds across three vectors—raw intelligence, intelligence per cost, and intelligence per cost at speed—driven by cost-effective models.

  • Charts to Understand AI: November 2026

    Charts on AI capacity, spending, adoption, and work, with historical comparisons, source notes, and data periods from Alephic’s November 2026 edition.

Related terms:

  • Inference

    Inference is the process of running a trained model on new input to generate a prediction or output—such as sending a prompt to GPT-4 and receiving a...

  • Transformer

    The transformer is the neural network architecture introduced in Vaswani et al.’s “Attention Is All You Need” that replaces recurrence with parallel...

  • WWGPTD

    WWGPTD began as internal Slack shorthand to remind teams that using AI isn’t cheating but the essential first step. It reframes strategy by asking how AI.

Token | Alephic