Agent Atlas

Atlas · models

Language model

A program trained on vast amounts of text to predict what comes next. Text goes in, text comes out.

Established practiceDepth 1 of 4Reviewed 2026-11-08

See it working: Plate I, Plate V

In plain words

A language model is a program trained on a huge amount of text. Give it text and it continues it: it answers a question, summarizes a page or writes code. Chat assistants and coding agents are built on one.

How it works

During training the model learns to predict the next word from the text before it. It works in tokens, pieces of words, not letters. To reply, it predicts one token, adds it to the text, and predicts the next, until it stops.

Going deeper

What the model knows about your conversation is only what is in its context window, which Anthropic’s glossary calls the model’s working memory. Everything else it knows comes from training. Plate I’s model is a scripted stand-in, labelled simulated. It shows where a model sits in an agent, not how one works inside.

Sources

  1. Glossary Anthropic
    • These models are trained on vast amounts of text data and can generate human-like text, answer questions, summarize information, and more.
    • pretrained to predict the next word, given the previous context of text in the document.
    • represents a “working memory” for the model.

Quotes were checked against the live pages on 2026-11-08.