Definition
Context window
A context window is the maximum amount of text, measured in tokens, that a model can process in one request. Current Claude Fable 5, Opus 5 and Sonnet 5 tiers support 1M tokens; GPT-5.6 Sol, Terra and Luna expose 1.05M through the API; Gemini 3.1 Pro supports 1M. Host products can expose smaller effective windows than the model allows.
Why it matters
Managing context is one of the most important skills in AI-assisted development. The published model limit is a ceiling, not an allowance: what the agent actually carries is the window minus the tools, the config files and everything already said in the session.
How each tool handles it
- Cursor
- Up to 1M tokens in Max Mode.
- Claude Code
- 1M on current Sonnet, Opus and Fable tiers.
- Codex
- Model- and surface-dependent; the GPT-5.6 API exposes 1.05M.
Frequently asked questions
Is a bigger context window always better?
Not on its own. A larger window raises the ceiling but does not decide what fills it, and filling it with everything is how relevant instructions get buried. What decides quality is which files, rules and history the agent is carrying.
Why is the effective window smaller than the model limit?
Because the host product decides how much of the model’s window it actually uses. Tool definitions, configuration files and prior turns all occupy the same budget before your prompt does.
What does compaction do to the context window?
Compaction summarises conversation history to free space, preserving key decisions while discarding verbose intermediate steps. Claude Code triggers it with /compact; in Codex it happens automatically as the context approaches the model's limit.