Why your Claude usage limit runs out so fast. A short question in a long chat isn't short

You have been in one Claude chat since the morning. First a client proposal, then a supplier contract pasted in as a PDF, then two versions of your terms of service to compare. In the afternoon you type a short "fix the typo in the second paragraph" and get a message that your Claude usage limit for this session is used up.

That isn't a glitch or bad luck. Most of the limit goes on what travels along with your question. The question itself is usually the small part.

How the Claude usage limit works

Paid Claude plans have two limits. The session limit resets every five hours, and weekly limits reset once a week. The chat in the browser, the desktop app and Claude Code all draw from the same pool.

You can see how much you've used in Settings, under Usage. There are bars for the current session and for the whole week. We covered the basics in our piece on choosing a model, so this one is about what actually moves those bars. Anthropic lists the length and complexity of the conversation, the features you use, the model and the effort level.

Every message carries the whole conversation

A model doesn't keep the chat in memory between messages. To answer, it works through the conversation from the start every time. Anthropic says this plainly in the Claude Code documentation. The full conversation goes with every request, so a one-line question in a session that has been open since the morning still uses limit for the whole conversation.

The same documentation mentions one more thing. After a longer break the cache expires, and your first message processes the whole context again. That's exactly the after-lunch moment from the start of this article.

The chat works in a similar way. Conversation length is on the list of things that affect usage.

When a chat gets very long and code execution is on, Claude summarises earlier messages on its own. According to Anthropic, those conversations use more of the limit, and it's better to start a new chat. We described how to carry your decisions over to a fresh chat in the article on long AI conversations.

The same files sent again and again

A large attachment counts for more than a short message. If you attach the same product catalogue or the same terms to every new chat, it uses the limit again each time.

Documents you use often are better kept in a project. Claude caches files added to a project, and they count for less when reused. The cache expires after a period of inactivity, so the first message after a long break counts the whole file again.

Pasting the whole document when you need one part also eats the limit. For a question about the notice period, the one clause about notice is enough. You don't need the full contract with all its annexes.

Connected tools and web search

Connectors to your calendar, email or drive are handy, but each one takes up room in the conversation. Anthropic's help pages say tools and connectors are token-intensive and recommend turning off the apps a conversation doesn't need. The same goes for web search. If you don't need current information, tell Claude not to search.

There's also a setting that controls when connectors load. Click the "+" button in the lower left of the chat, hover over Connectors, then Tool access. There are three modes to choose from:

• Auto is the default. Claude decides which connectors to load based on what you're working on.
• Always available loads all your connectors at the start of every conversation.
• On demand loads a connector only when Claude searches for the right one for your request.

Anthropic suggests switching to On demand if you have 10 or more connectors active. Research, running code, creating files and browsing websites are multi-step tasks, and they count toward the limit too.

If someone on your team works in Claude Code, the /usage command shows what share of recent usage went to subagents, plugins and individual MCP servers. It also suggests what to change.

The strongest model on small tasks

Higher effort gives more thorough answers, but it takes longer and uses more tokens. Anthropic recommends Low and Medium for routine work. Rewriting an email or a table on Max spends limit for nothing.

The strongest models have one more catch. On Pro, and on standard seats in Team and Enterprise plans, the Fable models aren't part of the plan's usage limits. They run on usage credits, extra usage that is billed separately. On Max they count toward the weekly limit, up to half of it, and use it up faster than other Claude models.

When the limit disappears faster than usual, open the Usage page, start a new chat with a short summary and switch off the connectors that conversation doesn't need. Keep the strongest model for tasks that really need thinking.