What is a context window?
A context window is the token budget a model can hold in one request. In a coding agent it is spent on the system prompt, CLAUDE.md or AGENTS.md, tool definitions, file reads, command output and conversation, all reread every turn. When it fills, the agent compacts older material and instructions it was following can quietly disappear. Its size depends on the model and the host.
- In Polish
- Okno kontekstu
- First set up at
- All levels, free guide
- Also searched as
- context length, token limit
$59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.
Why it matters.
A larger window postpones compaction; it does not make the model attend equally well to everything in it. Instructions from early in a very long session compete with hundreds of thousands of tokens of file contents and logs, so treat the window as a budget you spend on the task, not as storage.
What fills the window
The window is not only your conversation: every turn resends everything the agent currently holds, so the largest consumers are the ones you do not see in the terminal.
| What takes the space | Loaded when | How to shrink it |
|---|---|---|
| System prompt, environment info | Always, first | You cannot; it is fixed overhead |
CLAUDE.md / AGENTS.md and rules | At session start, and again after compaction | Prune the file; move reference material into skills or path-scoped rules |
| MCP tool definitions | At start or on demand, depending on the tool | Remove servers you do not use |
| File reads | Each time the agent opens a file | Name the exact files; ask for the one function, not the whole file |
| Command output | Each test, build or grep run | Filter output (--silent, pipe logs through tail -50); fail fast |
| Conversation history | Grows every turn | Compact with a focus, or clear between tasks |
Compact, clear, fork or delegate
Decide by what you still need, not by how full the meter is.
| Situation and action | Claude Code | Codex | Cursor |
|---|---|---|---|
| Same task, long history, decisions still matter: compact with a focus | /compact <what to keep> | /compact | Write the handoff note, then start a new chat |
| Only part of the history is noise: summarise a slice | /rewind, then Summarize from here or Summarize up to here | No slice summary in Codex CLI 0.157.1: write the handoff note, then /compact or /new | Write the handoff note, then start a new chat |
| The next task is unrelated: clear | /clear | /new | New chat |
| Try another approach without losing this one: branch | /branch | /fork | Start a new chat from the handoff note |
| A side task would flood the window with reads or logs: delegate | A subagent: its reads stay in its own window and only the summary returns | Subagents (/subagents switches between them) | A subagent (Cursor docs, checked 2026-08-28) |
| A quick question about the session: ask on the side | /btw <question>, which does not add to the conversation | /btw <question> (a side conversation in an ephemeral fork; Ctrl+C returns to the main thread) | Ask in a separate chat |
/clear and a new session; decisions made only in the conversation do not.How each tool does it.
Cursor, Claude Code and Codex each name this their own way. Each note carries the date it was checked.
Cursor
Check the model picker for the current window; subagents keep large exploratory reads out of the main chat.
The guide does not name Cursor’s context indicator or its summarisation thresholds, because they change by model.
Cursor docs, 2026-08-28
Claude Code
Opus 5.5, the default model from v2.1.280 (both release channels on 2026-10-12), has a 1M-token window on the Anthropic API; /context shows what fills it.
By default a session on a 1M-window model auto-compacts at about 967K tokens. /compact <what to keep> compacts on demand, and /autocompact 500k makes automatic compaction start earlier.
Context windows guide, 2026-09-26
Codex
A default working window of 272K tokens, raised with model_context_window up to 872K for the GPT-6 and GPT-5.6 models (the 0.157.1 catalog, and again the 0.162.1 catalog, checked 2026-10-12).
Input above 272K is billed at a higher tier, so raise the window only for the sessions that need it, for example with -c model_context_window=400000 on one run.
Models hub, Codex CLI 0.157.1, 2026-09-26
Questions about the context window.
Is a bigger context window always better?
No. A larger window postpones compaction, but it does not make the model attend equally well to everything in it. Instructions from early in a very long session compete with hundreds of thousands of tokens of file contents and logs, so treat the window as a budget you spend on the task, not as storage.
Why can the working window be smaller than the model’s limit?
Because the host decides how much to use. Codex gives its default model a working window of 272K tokens although the GPT-6 and GPT-5.6 models allow up to 872K, and a Claude Code session on a 1M-window model auto-compacts at about 967K tokens by default. The system prompt, instruction files and tool definitions take their share before your prompt does.
What does compaction do to my instructions?
It summarises older material to free space, so detail is dropped. In Claude Code the project-root CLAUDE.md is re-read from disk afterwards, subdirectory files and path-scoped rules come back when Claude next reads a matching file, and instructions you gave only in the chat are gone. Compact on demand with /compact, ideally before a phase change, and keep anything that must survive in a file.
How do I see what is in the window?
In Claude Code run /context: it draws usage by category (system prompt, memory files, tools, messages) and suggests savings. In Codex run /status for the session configuration and token usage. For Cursor, check the model picker for the current window.
Keep reading.
The guides that go deeper, and the terms and comparisons next to this one.
In the docs
Related terms
Does this page answer your questions?
One tap, no sign-in.
What are you here for?
What is holding you back?
How does it read?
Thanks. If you left an email, Michał will reply.
Answering your questions?
One tap each. Close whenever you like.
no sign-in
Read the guides in the same words.
Open every guide with the Toolkit. Each term is defined once here, and the guides use it the same way.
$59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.
Michał replies, not a bot. The address is used for this reply only.