Models and Tokens

What a model is

An AI model is a program trained on data to map inputs to outputs. It does not follow written rules; it predicts. A Large language model is the text version: it generates output by predicting one token at a time. In practice: GPT, Claude, and Gemini are all large language models behind an API.

Tokens

A Token is the unit a model reads and writes. It is usually a word fragment, around 3 to 4 characters of English. In practice: 1,000 tokens is roughly 750 English words.

Tokenization is the step that splits text into tokens before the model sees it. Different model families split text differently, so the same sentence can cost a different token count on different providers. In practice: code and non-English text often tokenize into more tokens than plain English.

The context window

The Context window is the maximum number of tokens a model can hold in one request: your instructions, the conversation so far, retrieved documents, and the answer it is writing. Anything beyond the limit gets cut or must be summarized. In practice: a 200K window fits roughly 150,000 words of combined input and output.

How pricing works

Providers separate Input vs output tokens. Input is everything you send; output is everything the model generates. Output usually costs several times more because generation is the expensive step. In practice: a long prompt with a short answer is often cheaper than a short prompt with a long answer.

Prices are quoted as Per-token pricing: dollars per one million tokens, with separate input and output rates. In practice: at $3 per 1M input and $15 per 1M output, a 1,000-token prompt with a 1,000-token answer costs about $0.018.

One more split: an Open-weight vs proprietary publishes its trained weights so anyone can self-host it, while a proprietary model is only reachable through the vendor's API. In practice: Llama and Mistral publish weights; GPT and Claude do not.

A detail that surprises people: a model can be retired from a consumer app and still be fully available in the API. Vendors keep API versions alive for the software built on them. Retired in ChatGPT does not mean retired in the API.

The Cromus Lens:

Cromus does not run models. It prices and governs the work that models will do. The Verified Model Registry tracks active models across a dozen providers with admin-verified per-1M pricing and context windows, and it distinguishes active, legacy, preview, and historical entries because API availability outlives consumer UI availability. When you simulate a workflow's cost in Cromus, every number comes from this registry, not from a guess.

Use the panel below to put two real registry entries side by side.