Claude and Claude Code glossary
The words you run into when talking about limits, models and Claude Code, each defined in a sentence or two, from the official documentation.
Limits and plans
- 5-hour session
- The rolling five-hour window on which a Claude plan’s short limit resets.
- Weekly limit
- The weekly cap, shared by all models, that resets at a fixed day and time assigned to each account.
- Max 5x, Max 20x
- The two tiers of the Max plan: 5 or 20 times Pro’s usage per session.
- Usage credits
- A prepaid balance that takes over when a limit is reached, so you can keep going without changing plans.
- Reset
- The moment a counter starts over: every 5 hours for the session, every week for the weekly limit.
Models and tokens
- Token
- The unit of text a model reads and writes; the API bills per million tokens. According to Anthropic, Claude 4.7 and later models split the same text into about 30% more tokens.
- Input, output
- The tokens sent to the model and the ones it produces. Output costs five times as much as input on the Claude API.
- Prompt cache
- Part of the context kept server-side: re-reading it costs a fraction of the input price. It lasts 1 hour on a subscription.
- Effort
- The setting that tells the model how much to think. Lowering it saves your limits, often better than switching models.
- Batch processing
- API requests sent in bulk and processed later, billed at half price.
Claude Code
- Context window
- Everything the model sees at a given moment: conversation, files read, instructions. Fable 5.1, Sonnet 5.5 and Opus 4.7 and later natively have one million tokens.
- Compaction
- A summary of the conversation that frees up context. Claude Code does it on its own near the limit; /compact triggers it by hand.
- /clear
- Starts over with an empty conversation. Unlike /compact, it costs nothing.
- /context
- Shows how full the context is, as a colored grid.
- /usage
- Shows the session cost, your plan’s limits and activity stats.
- CLAUDE.md
- The instructions file Claude Code loads at startup. Anthropic recommends staying under 200 lines.
- Subagent
- A secondary agent that works in its own context and only returns its result: useful for verbose tasks.
Frequently asked questions
What’s the difference between the session limit and the context window?
The session limit covers your total usage over five hours; the context window, what the model can see in a single conversation.
Does compaction use up my limits?
Yes: according to the documentation, /compact is itself a large request. /clear costs nothing.