Claude is great, but it can burn through tokens fast if you aren’t careful. If you’re reading this post, then…you probably haven’t been careful enough. How naughty.
The good news? Most of you probably don’t need the $100 Claude plan. You just need to stop wasting tokens like a maniac.
Claude probably doesn’t want you to read this, so here it is.
These tips apply mostly to Claude Chat, Claude Projects, and Claude Cowork workflows. Some are Cowork-specific, while others are general Claude habits that help you burn fewer tokens and avoid wasting usage. Some can even be helpful with Codex or OpenClaw.
Tip #1: Shrink Your Files (No PDFs)
PDFs are token whales.
A 15-page PDF can turn into around 45,000 tokens inside Claude, while that same text saved as a markdown file can be closer to 2,000 tokens.
That is not a small difference. That is the difference between “I’m using AI efficiently” and “I accidentally fed Claude my last box of Hot Pockets.”
Here’s what you do:
Open a Google Doc. Paste the text from your PDF. Download it as an .md file. Upload the markdown file instead.
Now you’re using the same information with way fewer tokens.
Tip #2: Plan First. Build Last.
Do not ask Claude to build the whole thing immediately.
First, open Claude Chat and prompt:
Help me plan a [financial model for my business]. Ask me questions first.
Once Claude gives you the final plan, paste that plan into Cowork and say:
Then paste the final plan into Claude Cowork when the task requires building, organizing, researching, or working across files, and prompt, “Build this.”
That one step can save a ton of usage because Claude is no longer guessing, rambling, rebuilding, correcting itself, and doing its usual “I am thinking very deeply” performance.
Always remember this: plan first with Claude Chat, then build.
Tired of AI filters and data-harvesting in tools like Claude? Try Venice today, built for more creative freedom and privacy. Get 20% off any paid Venice subscription for a limited time with promo code RUNTHE20. Disclosure: This is an affiliate link, and I may earn a commission if you purchase.
Tip #3: Let Claude Ask You Questions
Stop writing huge prompts when Claude can ask you for the missing details.
Instead of writing 1,000 words explaining every detail, use this:
I want to [TASK] to [SUCCESS CRITERIA]. Ask me questions, one at a time, using AskUserQuestion before you start. Do not ask more than one question at a time.
Then select the answers.
Clicking answers costs almost nothing. A 500-word prompt costs 1,000 words’ worth of tokens, which is a silly way to burn usage before the real work even begins.
Simple math, painful lesson.
Tip #4: Edit Your Messages, Don’t Resend
This one is where people quietly torch their credits.
If you make a mistake, do not send:
No wait, I meant…
Click Edit on your last message. Fix the mistake. Hit Save.
Claude regenerates from the corrected version instead of stacking a new message on top of the old one.
Every “oops, actually…” makes the conversation heavier, and Claude has to carry all of it like the weight of this sick, sad world.
Tip #5: Summarize Every 15 Messages
Long chats get expensive because Claude keeps dragging more and more context forward.
After about 15 messages, ask Claude:
Summarize this entire conversation into a brief.
Copy that summary. Start a new chat. Paste it as the first message.
You just compressed a giant conversation into a small brief, which means Claude can work from the important stuff instead of rereading the entire saga like it’s studying for finals.
Instead of hauling 105,000 tokens around, you might only need 500 now. Congrats.
Tip #6: Switch Claude Models Before You Start
Do not use Claude Opus for everything.
Before starting, click the model dropdown and choose the model that matches the job. If the task should take Claude under 30 seconds, use Haiku or Sonnet. Save Opus for deep, multi-step work where you actually need the extra reasoning.
This could save you 3–5x per message.
Tip #7: Use Claude Projects, Not Uploads
If you upload the same file into five different chats, Claude has to process it again and again.
Use Projects instead.
Go to Projects. Create one. Upload your files there once. Then start new chats inside that Project.
Claude can reference the files without you constantly re-uploading the same contract, document, guide, or cursed client PDF.
Tip #8: Turn Off Extra Features in Claude
Extra tools can drain usage, especially when they are sitting there enabled for no good reason.
Go to the tools panel and turn off what you do not need. Things like:
Web Search
Connectors
Extended Thinking
Only turn on the tools needed for that specific task.
Do not leave everything running in the background like the guy at work with ADD who has 47 Chrome tabs open.
Tip #9: Batch 3 Tasks Into 1 Message
Do not send three separate messages like this:
Summarize this.
List the key points.
Write a headline.
Send one message:
Summarize this, list the key points, and write a headline.
Three messages mean three context reloads, while one message means one reload.
Same work, less waste.
Wrapping It Up
You probably don’t need the $100 Claude plan, or at least most of you don’t.
To save tokens in Claude, you need to shrink your files, plan before building, let Claude ask questions, edit messages instead of resending, summarize long chats, use the right model, rely on Projects, turn off extra tools, and batch tasks together.
None of these tricks are complicated, which is the weird part. They’re free, easy, and sitting right there while half the internet complains that Claude is too expensive.
Try these fixes before upgrading. Your credits will last longer, Claude will feel less stingy, and you can spend the extra money on something more useful, like beer, hard drives, or donuts.
Now that you’re saving tokens and money with Claude, check out our guide on how to save money with OpenClaw.
Add a comment in the comment section below and let me know which Claude usage trick saved you the most tokens.
Until next time, remember to run the prompts and prompt the planet.
