Stop burning through Claude's limit. Here's how.
Hitting your Claude limit before lunch is a sign of avoidable waste. Here's where your tokens are going and the habits that fix it.
Liked this? New guides land here first. No spam, ever.
Claude reads the entire conversation from scratch every time you send a message. By message 20, you're paying for messages 1 to 19 all over again. That one fact explains most token problems. The fix is a handful of small habits: fresh chats for new topics, files stored in Projects, the right model for the job, and batched questions instead of long back-and-forth threads. This guide covers all of them.
What a token actually is.
Tokens are the fuel Claude runs on. Every message you send, every file it reads, and every response it writes all cost tokens. One token is roughly four characters, which works out to about three quarters of a word in English. A thousand tokens is somewhere around 750 words, or two to three pages of text.
Here's the part most people don't know. Every time you send a message, Claude doesn't just read your latest prompt. It rereads the entire conversation from the beginning to understand the context. So message 20 costs significantly more than message 1, because you're paying for all 19 previous exchanges again. Every single time.
That one insight changes how you should use Claude. And it's the reason most of the tips below work.
1. Start a new chat for every new topic.
This is the single biggest one. Long chats are a silent drain. A 50-message thread is 50 times more expensive than the first message you sent. Every new topic deserves a fresh chat, because starting over brings your token cost back to zero.
If you're worried about losing context when you close a thread, ask Claude to write you a handoff summary first. Then paste that summary into the new chat and carry on.
Before closing a long chat: "Summarise everything we've covered in this conversation. Include the key decisions made, what I asked you to do, and any context I'll need to continue this work in a new chat."
2. Store files in Projects.
If you paste the same PDF or document into multiple chats, Claude counts those tokens every single time. Projects fix this. Drop your files into a Project once, and Claude can access them in every conversation within that Project without reprocessing them from scratch.
Set up a different Project for each area of your work. One for client work, one for your content, one for admin. The files you drop in there don't count against your usage limit the same way repeated pastes do.
3. Edit instead of following up.
When Claude gives you a response that's not quite right, most people type a follow-up message. That follow-up forces Claude to reread the whole thread again before responding. There's a better option: click the pencil icon on your original message, rewrite it, and let Claude regenerate. The bad response disappears as if it never happened, and you haven't added any extra weight to the thread.
One edited message beats two messages every time.
4. Match the model to the task.
Claude has three main models: Haiku, Sonnet, and Opus. They don't all cost the same. Opus burns through your quota three to five times faster than Sonnet. Using Opus for everything is like driving a lorry to the corner shop. You'll get there, but you'll burn far more fuel than the trip needed.
Use Haiku for simple tasks: grammar checks, quick questions, short formatting jobs. Sonnet is your everyday model for writing and most work tasks. Save Opus for the genuinely complex stuff, hard decisions, deep analysis, work where the extra reasoning actually matters. You can switch using the model selector at the bottom of the chat.
5. Batch your questions.
Three messages means three full context loads. One message with three questions means one. If you have several things to ask, list them together in a single message. Bullet points work well for this. It's a small habit that adds up fast over a day of heavy use.
Instead of: "Can you check this?" / "Actually also fix the tone" / "And make it shorter" — send one message: "Check this for errors, adjust the tone to be warmer, and cut it to 150 words."
6. Be specific in your first message.
"Help me write a blog post" kicks off a back-and-forth that costs you several exchanges before you get anything usable. "Write 500 words about AI tools for new freelancers, conversational tone, end with a practical next step" gets you a usable draft in one shot. The more specific your first message, the less you need to go back and forth.
This is especially true for content: include the word count, the audience, the tone, and the format in your opening prompt. You'll get fewer rounds of iteration, which means fewer tokens spent.
7. Use Plan Mode before you ask for code.
If you're using Claude for code, use Plan Mode first before it starts writing anything. Reviewing a text plan costs almost no tokens. Correcting half-written code costs a lot. Get Claude to agree on the approach first, confirm it understands what you want, and then ask it to build it. One round of correction in plan form is cheaper than three rounds of corrections on actual code.
8. Use /compact or /clear in the Code tab.
If you're working in Claude's Code tab, two commands are worth knowing. Typing /clear wipes the session history and starts fresh. Typing /compact does something smarter: it summarises everything so far into a checkpoint, keeping the essential context without carrying the full weight of every exchange. If you're deep into a long coding session, /compact is your friend. These commands only work in the Code tab.
9. Set up a claude.md file for repeat projects.
If you work on the same project regularly, you're probably repeating the same context at the start of every session. Your preferences, your rules, your setup. A claude.md file fixes this. It's a plain text file you drop into a project folder. Claude reads it once at the start of every session, so you never have to type that context again. Over weeks of regular use, the token saving is significant.
10. Time your sessions across the day.
Claude's usage limits work on a rolling five-hour window, not a daily reset at midnight. The window starts from your first message. So if you start at 7am, that window closes at noon. Start a new session at noon and you open a second window. One at 5pm gives you a third.
If you do all your heavy work in a single two-hour block, you're burning through one window fast. Spread it out across the day and you're much less likely to hit a limit. And the simple rule: start early. The earlier your first session, the more complete windows you can fit in before you finish for the day.
11. Turn off tools you're not using.
Web search, code execution, and connected apps all burn tokens behind the scenes, even when you don't explicitly ask for them. If you're not using them for a particular session, switch them off. There's a tools toggle in the chat interface. It's a small thing, but it removes silent overhead from tasks that don't need it.
One more thing about Co-work.
If you use Claude's Co-work tab, know that it is the heaviest tab by far. To verify it's doing things correctly, Co-work regularly takes screenshots of your virtual desktop, and images are expensive. A single screenshot costs around 1,300 tokens on its own. Multiply that across a long task and it adds up fast. Co-work is worth using when a task genuinely needs it. Just be aware of what it costs so you use it intentionally rather than by default.
If you're on the Pro plan, token limits only become a real issue when you're running Co-work heavily or doing long Code sessions. For most standard chat use, good habits will keep you well within your limit. Match your plan to your actual usage, not to what sounds impressive.