Never hit Claudes Usage Limit Again

Did you know that your 10th message to Claude cost 11 times more than your first? Not because Anthropic is charging

first? Not because Anthropic is charging you more, it's because Claude re-reads your entire chat history before replying to the message that you just sent. I

used to hit Claude's usage limits every single day. However, after implementing

single day. However, after implementing a few practical rules for how I use Claude, I haven't hit my limits in over a week. So, without further ado, here

a week. So, without further ado, here are the 11 things I changed that stopped me from hitting Claude's usage limits.

Oh, and if we haven't met yet, I'm Duby.

I build apps every day using Claude and I've generated 47 thousand dollars in less than 50 days using these AI tools.

So, first thing I need you to do in order to save your precious Claude usage is you need to stop looking at it as how many messages you sent and instead look at how many tokens you use. Because

Claude doesn't count messages, it counts tokens. A token is roughly 3/4 of a

tokens. A token is roughly 3/4 of a word, so a 100-word message is about 130 tokens. Every time you send a message,

tokens. Every time you send a message, Claude re-reads the entire conversation from the top.

So, your first message costs maybe 500 tokens, your second message costs 1,000, your 10th message costs 5,000, and your 30th? That's 232,000 tokens for one

30th? That's 232,000 tokens for one message. A developer named Aniket

message. A developer named Aniket Parihar actually tracked this. 98.5% of

his tokens went to re-reading old conversation history, only 1.5% went to generating the actual response. That

means for every 100 tokens you're burning, only one and a half are doing useful work. That's not Claude being

useful work. That's not Claude being stingy, that's you feeding it a novel every time you ask a question. Once you

understand this, everything will change because every single rule I'm about to give you comes back to this one idea: use fewer tokens per interaction. And

just like the tokens you put in count against you, the inverse is also true.

Every word Claude sends back counts as well. So, check out this repo called

well. So, check out this repo called Caveman. It makes Claude drop the fluff

Caveman. It makes Claude drop the fluff and talk, well, like a caveman. People

have reported token reductions of up to 65% from using this alone. I'll leave a link in the description. Rule one, don't follow up. This is the one that saved me

follow up. This is the one that saved me the most tokens immediately. When Claude

gets something wrong, your instinct is to send a correction. No, I meant this.

Uh, that's not what I wanted. Try this

instead. Don't do that. Every subsequent

message gets added to the conversation history and remember, Claude re-reads all of it every single turn. So, you're

burning tokens on context that didn't even help. At five messages deep, you're

even help. At five messages deep, you're already at 7,500 tokens. So, instead of correcting Claude with a follow-up, click edit on your original message, fix the wording, hit regenerate, the old

exchange gets replaced instead of multiplied. Exact same result, but for a

multiplied. Exact same result, but for a fraction of the cost. Rule number two, a fresh chat every 15 to 20 messages.

Imagine a chat with 100-plus messages.

That's over 2.5 million tokens burned.

Most of it just re-reading old history.

So, ideally, start a new chat every 15 to 20 messages. When a conversation gets long, ask Claude to summarize everything. Copy that summary, start a

everything. Copy that summary, start a new chat, and paste it as your first message. You keep the context, you ditch

message. You keep the context, you ditch the bloat. I started doing this and my

the bloat. I started doing this and my limit warnings basically disappeared overnight. Rule number three, batch your

overnight. Rule number three, batch your questions into one message. Many people

believe that splitting questions into separate messages leads to a better result. Almost always, the opposite is

result. Almost always, the opposite is true. Three separate prompts equals

true. Three separate prompts equals three context loads. One prompt with three tasks equals one context load. You

save tokens twice. Fewer context reloads and you stay further from hitting your limit. Instead of summarize this

limit. Instead of summarize this article, now list the main point, now suggest a headline sent separately, instead just write summarize this article, list the main points, and suggest a headline. And guess what? The

answers often turn out better because Claude immediately sees the full picture. Three questions, one prompt,

picture. Three questions, one prompt, always. Four, track your token usage.

always. Four, track your token usage.

[music] Claude doesn't show you how many tokens you're actually using. Just a progress bar that says 63% used. That's it. So, a

developer called Paweł Huryn built a local dashboard that reads files Claude Code already writes to your machine.

Turns out, every session, every turn, every single token is logged to your machine in JSONL files. Input tokens,

output tokens, cache reads, cache creation, model names, timestamps, it's all there. You just can't see it. The

all there. You just can't see it. The

dashboard scans those local files, builds a database, and serves charts on localhost. You can filter by model,

localhost. You can filter by model, filter by time range, cost estimates based on current API pricing, works retroactively on your entire Claude Code history. It's free, open source, zero

history. It's free, open source, zero dependencies, you just need Python. You

can't fix what you can't measure.

Install it, look at your actual numbers, and I guarantee you'll be shocked at where your tokens are actually going.

Rule number five, upload recurring files to projects. If you upload the same PDF

to projects. If you upload the same PDF to multiple chats, Claude re-tokenizes that document every single time. Use the

projects feature instead. Upload your

file once, it gets cached. Every new

conversation inside that project references without burning tokens again.

Cached project content doesn't eat into your usage when you access it repeatedly. If you work with contracts,

repeatedly. If you work with contracts, briefs, style guides, or any long documents, this alone could cut your token spend dramatically. Rule number

six, set up memory and user preferences.

Every new chat without saved context wastes three to five messages on setup.

I'm a marketer, I write in a casual style. I prefer short paragraphs. You've

style. I prefer short paragraphs. You've

probably seen people start every prompt with act as a professional whatever.

That's token burned on repeat. Claude

can actually remember this permanently.

Go to settings, then memory and user settings, save your role, communication style, and settings once. Claude will

automatically apply them to every new chat. Rule number seven, turn off

chat. Rule number seven, turn off features you're not actually using.

Things like web search, connectors, and explore mode, all of these add tokens to every response, even if you don't use them. For example, if you're writing

them. For example, if you're writing your own content, turn off the search and tools feature. The advanced thinking feature also consumes tokens. Keep it

turned off by default. Only turn it on if your first attempt was unsatisfactory. The rule of thumb is if

unsatisfactory. The rule of thumb is if you didn't turn this feature on intentionally, just turn it off. Eight,

use Haiku for simple tasks. Grammar

checking brainstorming formatting quick translations, short answers, Haiku handles all of this at a much lower cost than Sonnet or Opus. Choosing the right model is the most important decision you

make every single day. Haiku for drafts and simple tasks frees up 50 to 70% of your budget for tasks that truly require powerful models. The way that I look at

powerful models. The way that I look at it is Haiku is for quick tasks, it's low cost. Sonnet is for real work, medium

cost. Sonnet is for real work, medium cost. Opus is for deep thinking and high

cost. Opus is for deep thinking and high cost. You don't need powerful models for

cost. You don't need powerful models for simple tasks. Rule nine, and this one

simple tasks. Rule nine, and this one might sound obvious, but spread your work across the day. The Claude system uses a rolling 5-hour window. It does

not reset at midnight. Your limit

gradually decreases. Messages sent at 9:00 a.m. will no longer count by 2:00

9:00 a.m. will no longer count by 2:00 p.m. If you use up your entire limit

p.m. If you use up your entire limit during a single morning session, most of your daily limit will remain unused.

Divide your day into two to three sessions morning afternoon and evening. By the time you return, your

evening. By the time you return, your previous usage is no longer counted and you have a new limit. Quick cheat code, set a cron job to literally just ping Claude with one tiny message at 6:00

a.m. while you're still in bed. Your

a.m. while you're still in bed. Your

5-hour window starts then, not when you sit down to work. So, if you hit your limit at 10:00 a.m., your reset lands at 11:00. 10, work during off-peak hours.

11:00. 10, work during off-peak hours.

Starting March 26, 2026, Anthropic will now use up your 5-hour session limit more quickly during peak hours. 5:00

a.m. to 11:00 a.m. Pacific time, 8:00 a.m. to 2:00 p.m. Eastern time on

a.m. to 2:00 p.m. Eastern time on weekdays. Same query, same chat, but

weekdays. Same query, same chat, but during peak hours, it impacts your limit more. Your weekly limit remains the

more. Your weekly limit remains the same, but how it's distributed has changed. Running resource-intensive

changed. Running resource-intensive tasks in the evening or on weekends will significantly stretch your plan. If

you're outside the US and Europe, Latin America, or Asia, peak hours may actually fall during your afternoon, so check the calculation based on time zones. If you're like me and you live in

zones. If you're like me and you live in Australia, your off-peak hours are actually in the morning and in the midday. That is usually when you can use

midday. That is usually when you can use Claude without any worries. Rule 11,

enable extra usage as a safety net.

Subscribers to the Pro or Max plan can enable the overage feature in the settings usage section. When your

session limit is reached, Claude won't block your access. It will switch to pay-as-you-go billing at API rates. You

set a monthly spending limit to avoid unexpected bills. This isn't about

unexpected bills. This isn't about saving tokens, it's about not losing your work at the worst possible moment.

At first, it will be very difficult to follow all the rules, but once you can apply them automatically, you'll almost never hit your limits. You might even switch from the Max plan to a regular one. You'll have plenty of tokens to

one. You'll have plenty of tokens to use. Now, whenever you see your usage

use. Now, whenever you see your usage limits go up, just remember, Claude doesn't count your messages, it counts your tokens.