You pay $20/month for Pro or $100 for Max or $200 for Max 20x.
And 75% of that money goes nowhere. Not to your tasks - to Claude talking to itself. And you let it do that.
I tracked my usage for a week. Most tokens were just burned into the void. Here's what I changed to save roughly $500 a month.
Part 1 - Teach Claude to talk like a caveman
Regular Claude on a web search task: ~180 tokens. It says "I'll help you search for that!" then searches, then explains what it found, then summarizes, then asks if you need anything else.
Caveman Claude on the same task: ~45 tokens. "Tool work , result, me stop."
135 tokens × 20 tasks a day = 2,700 saved tokens every day. Per month that's ~81,000 tokens you're no longer burning.
caveman - one skill that cuts 65-87% of token costs (1,900+ stars)
github.com/JuliusBrussee/caveman
Real benchmarks from the API:
Average across 10 tasks: 75% savings.
Three intensity levels:
lite -> removes fluff, keeps grammar (professional, no filler)
full -> removes articles, fragments (full grunt mode) ultra -> maximum compression (telegraph style, cuts everything)
Start with lite. Switch to full for agentic tasks where you're not reading the responses anyway.
Why it works: caveman Claude doesn't explain itself - it just does the task. Gives the result then stops. No "I'd be happy to help." minimum fluff
Result: a clear short answer to your question that saves both time and tokens
Replaces: Manually writing "be concise" in every prompt.
how did i find this here: https://t.me/noisyclub01
Part 2 - Stop losing 2-4 hours every morning
You start working at 8:30am and already hit the limit by 11, blocked until 1:00pm. Two dead hours in the middle of your workday.
you can game it very easily if you know the timings
How the window actually works
Claude usage runs on a sliding 5-hour window. The window starts when you send your first message, rounded to the hour.
First message at 8:30 -> window anchors at 8:00 -> runs until 13:00
Hit limit at 11:00 -> wait until 13:00 -> 2 dead hours
But if you send a throwaway "hi" via Haiku at 6:15am before starting work:
"hi" at 6:15 -> window anchors at 6:00 -> runs until 11:00 At 11:00 window resets immediately Next window: 11:00 - 16:00 -> zero downtime
Same total budget. Just distributed better across the day.
claude-warmup - automatic window anchor
github.com/vdesimone/claude-warmup
GitHub Actions cron job that sends "hi" to Haiku every morning. Fork it, add your OAuth token, set the schedule, done.
Works on Pro, Max 5x, Max 20x.
Don't want a repo? Use Claude's built-in scheduled tasks:
Replaces: Losing 2 hours every morning waiting for the window to reset.
Part 3 - Other leaks you're probably ignoring
Edit, don't reply
When Claude misunderstands, most people write "No, I meant X." Every new message stacks on top of all previous ones. Claude re-reads the entire history every time.
Hit Edit on your original message, fix it, regenerate. The bad exchange gets replaced not stacked.
Batch your questions
Three separate prompts = three full context loads. One prompt with three questions = one load.
Instead of:
Write:
Fewer reloads. Further from the limit. Answers are usually better - Claude sees the full picture at once.
Upload files to Projects once
Uploading the same PDF to multiple chats? Claude re-tokenizes it every time.
Projects: upload once -> cached. Every conversation inside the project references it without burning tokens.
100-page PDF = ~75,000 tokens.
Uploaded 5 times = 375,000 tokens burned. In Projects - 75,000 once, the rest is free.
Use Projects. Upload once -> cached. Contracts, briefs, style guides - this alone can save $15-40/month.
Save your context in Memory
Every new chat without saved context = 3-5 setup messages. "I'm a developer, I want short answers, always use TypeScript..."
5 messages × 500 tokens × 10 new chats a day = 25,000 tokens/day repeating the same thing.
Go to Settings -> Memory. Save your role and preferences once. Claude applies them to every new chat automatically. Zero setup tokens.
Full cheatsheet
Total: up to $500+/month in leaks you can plug in 15 minutes.
One more thing: if you're not in the US - peak hours (5:00 - 11:00 Pacific) burn through limits faster since March 2026. Run heavy tasks in the evening or on weekends - your plan stretches much further.
Claude doesn't count messages. It counts tokens. Stop giving it reasons to use more.
**You build your own life - so choose the right path. / If this was useful - follow /
more info in my tg channel: **https://t.me/noisyclub01
