10 ways to make your Claude limit last

Why you hit the limit

Anthropic describes the context window like this:

A chat works like an email thread. Each reply carries the whole thread below it, and Claude reads all of it before it answers. Anthropic’s documentation says each turn’s input “Contains all previous conversation history plus the current user message”.

So every message, file, search result and answer stays in the chat and is read again with every new message. That’s why long chats, big files and extra tools use up your limit fastest.

Content Claude has already read can be cached, and cached content counts less. But caches “expire after a period of inactivity”, so after a break it counts in full again.

Official reference: Usage and length limits

Official reference: Context windows

1. Choose the right model

Stop using Opus for everything. Anthropic’s own advice is to start lighter:

Official reference: Usage credits for paid plans

Which model for which job

“Uses your limit” comes from Anthropic’s model guide, which ranks Haiku “Lightest”, Sonnet “Moderate”, Opus “Heavy” and Fable “Heaviest”. The guide adds: “if you use Opus or Fable on a task that Sonnet or Haiku could handle, you may be using more of your limit unnecessarily”.

API prices are the standard rates on Anthropic’s pricing page, checked 14 October 2026. Haiku’s are for prompts up to 100,000 tokens. On a Pro or Max plan you don’t pay per token: there, the “uses your limit” ranking is the one that matters.

For Sonnet, Anthropic’s guide lists “Writing and content creation” and “Most "general purpose" work where you're not sure which model to use”. For Opus: “Deep research and analysis you'll question, redirect, and build on as you go”.

Scheduled tasks too

When you create a scheduled task, you choose “The model you want to use”. A task that runs every day uses your limit every day, so pick Haiku or Sonnet unless it really needs Opus. Anthropic lists Haiku for “High-volume, lightweight automation tasks”.

Official reference: Choosing the right Claude model

Official reference: Claude Haiku

Official reference: Claude pricing

Official reference: Schedule recurring tasks

Official reference: Enterprise consumption guide

2. New topic, new chat

Official reference: Usage credits for paid plans

A long chat makes every new message carry everything before it. Very long chats also trigger automatic summarizing, and Anthropic says “Longer conversations that trigger automatic context management consume more of your usage limit.”

Do this: when the topic changes, click New chat. If you need something from an older chat, copy just that part.

Official reference: Usage and length limits

3. Specify the output format

A quick answer? A slide? A 10-page report? Say which one.

Official reference: Get started with Claude Cowork

Chat and Cowork are merging into one conversation, rolling out starting with Pro and Max: “Ask Claude for what you need, and it can decide which tool to use”. That’s handy, but files cost more: “creating files will use more of your limit compared to normal chats with Claude”.

Do this: end your request with the format you want: “just answer here”, “an Excel with one tab per region”, “a 5-slide deck”, “a chart”, “an artifact”.

Official reference: Cowork and chat are one Claude

Official reference: Create and edit files with Claude

4. Stop re-uploading. Use a Project

If you upload the same brief or report to chat after chat, it counts in full every time. Put it in a Project once instead.

Anthropic sums it up: “This means you can work with the same documents repeatedly without using up your messages as quickly.”

How big is a document? A PDF page “typically uses 1,500–3,000 tokens”, so a 50-page report can be 150,000 tokens. Uploaded to a new chat, it counts in full every time.

One limit: “Caches expire after a period of inactivity. If you come back to a project after a long break, your first message counts that content in full again.”

Do this: open Projects, create a project, and add your brand guide, client briefs or recurring reports to its knowledge. Then start those chats inside the project.

Official reference: Usage limit best practices

Official reference: PDF support

Official reference: Create and manage projects

5. Spend a little more time on your prompt

Every follow-up message makes Claude read the whole chat again. Anthropic’s checklist, from the same page:

• “Is there any background information you can provide upfront?”
• “Avoid vague queries that may require clarification.”
• “If you have multiple related tasks or questions, group them in a single message.”
• “Take a moment to review your message for clarity and completeness to reduce the need for follow-up messages.”

Do this: not “shorter”, then “more formal”, then “add a subject line”. Send one message with all three, and say the format you want back.

Rewrite the email below.

Make it shorter and more formal, and add a subject line.
Context: it goes to [who], who needs to [decide or know what].
Answer here in the chat, no file.

Email:
[paste your email]

Official reference: Usage limit best practices

5 more tips that didn’t fit in the video

6. Paste the part you need, not the whole document

The whole text lands in the chat and is carried with every later message. Do this: paste only the section you need, or ask for the key points.

Official reference: Enable and use web search

7. Use Research only when you need many sources

Research “runs on Claude Opus 5.5 and falls back to Claude Opus 4.8 when needed. It uses these models no matter which model you've selected for your chat.” So choosing a lighter model doesn’t make it cheaper. Do this: for one quick fact, just ask. Turn on Research only for real research.

Official reference: Use Research on Claude

8. Turn off connected apps you don’t need

Anthropic explains why: “Tools and connectors are token-intensive.” Do this: “Using the "Search and tools" menu, disable any tools that aren't relevant to the current conversation or that you don't want Claude to be able to invoke.”

Official reference: Usage and length limits

Official reference: Custom connectors

9. Lower the effort for routine tasks

Higher effort responses “take longer and use more tokens, so you'll reach your usage limits faster.” Do this: open the model menu next to Send and check the effort. Keep it on the default or lower for everyday work, and save the higher settings for your hardest problems.

Official reference: Model, effort and thinking settings

10. Tell Claude when you don’t need the web

Anthropic: “Usage of web search and web fetch counts toward your usage limits.” In the merged Claude, “There's no web search toggle. Claude searches the web when it helps.” Do this: say it in your message, for example “don’t search the web, just use what I gave you.”

Official reference: Usage and length limits

Official reference: Enable and use web search

Official reference: Cowork and chat are one Claude

Not on this list, and why

“Turn extended thinking off.” It can’t be done on today’s models. Anthropic: “Thinking cannot be turned off in Claude when using Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1, or Claude Opus 5.”

Checking Settings → Usage. It shows where you stand, but it doesn’t save anything on its own.

Official reference: Model, effort and thinking settings