Review Context Window

A context window is the maximum number of tokens a language model can use in a single request. That total can include input, output, and, for some models, reasoning tokens. It limits what can be available during that request; it is not the model's training data or long-term memory.

Think of the context window as the size of the model's worktable. The active context is the material the application actually makes available on that worktable for a request. Depending on the system, it may include instructions, recent messages, documents, tool results, images represented for the model, and other supplied material. The window has a limit. When material will not fit, a system may reject the request, omit or truncate material, summarize or compact it, or select only relevant information. The exact behavior depends on the system.

A conversation can continue longer than the material the model can consider at one time. This helps explain why a system may appear to forget, overlook, or lose access to something from earlier. It also explains why more context is not automatically better: irrelevant or conflicting material can make the useful signal harder to find.

We may have a long conversation containing instructions, examples, corrections, and source information. As more information is added, some earlier content may no longer be included in the model's active input. We may need to restate the goal or provide a reviewed summary.

A context window is not the same as permanent memory, the complete conversation archive, or everything the model learned during training. Systems also handle context limits differently. Older material is not always simply pushed out in first-in, first-out order.

How much the model can consider at one time.

* [[prompt|Prompt]] * [[input|Input]] * [[output|Output]] * [[process-context|Process Context]]

* [Anthropic, “Context windows”](https://platform.claude.com/docs/en/build-with-claude/context-windows) — definition, token limits, distinction from training data, and limits of larger context. * [OpenAI, “Conversation state”](https://developers.openai.com/api/docs/guides/conversation-state) — context-window capacity, token accounting, and context management across requests.

* [Context Engineering](../../../../modules/context-engineering/module.md) * [Active Context Checkpoint](../../../cheatsheets/context-engineering/active-context-checkpoint.md)