🎯 In one sentence: Claude measures usage in tokens (not messages), resets on a rolling 5-hour window plus a weekly cap, and certain models/features burn through your limit faster than others.

What are tokens?

Tokens are how AI measures text. They're the fundamental unit of everything — your input, Claude's output, uploaded files, conversation history. Everything gets converted to tokens.

A rough guide:

• 1 token ≈ 4 characters or ¾ of a word
• 100 tokens ≈ 75 words
• 1,000 tokens ≈ 750 words (about 1.5 pages)
• A single PDF page ≈ 1,500–3,000 tokens
• A single image ≈ 1,200–1,600 tokens

When you send a message, Claude counts tokens for: your message, the entire conversation history (every previous message gets re-read), any uploaded files, any system instructions, and Claude's response. This is why long conversations cost more — the whole history gets processed every single time.

How the usage limit system works

Claude doesn't give you a fixed number of messages. Instead, it tracks your total token usage across two windows:

The 5-hour rolling window

Your primary limit resets every 5 hours from your first message. If you start chatting at 9 AM, your window resets at 2 PM. The number of messages you get depends on:

• Which model you're using (Opus uses the most, Haiku the least)
• How long your messages and responses are
• What files you've uploaded (PDFs, images, etc.)
• What features are active (extended thinking, web search, Research mode, MCP connectors)