Stop Hitting Your Claude Limit

Don't pay for $100/month. These 12 tips can save your tokens.

Before You Type

1. Convert Files Before Uploading

One PDF page = 1,500–3,000 tokens. Extract the text into a .md file instead.

• Wastes tokens: Raw PDF upload
• Saves tokens: .md text extract

2. Pick the Right Model

• Haiku for quick tasks (LOW COST)
• Sonnet for writing & coding (MEDIUM)
• Opus for deep reasoning (HIGH COST)

Use Haiku for drafts. Sonnet for real work. Opus for the hard stuff.

3. Turn Off Features You Don't Need

Web search, connectors, and Extended Thinking all add tokens to every response. Default everything off. Turn on per task.

4. Plan in Chat, Build in Cowork

Do your thinking in Chat (cheap), then build the final output in Cowork (expensive).

• Chat = Plan & think (Low cost)
• Cowork = Build output (Higher cost)

During the Chat

5. Say "Ask Me Questions"

A 15-word prompt + clicking options costs almost nothing. A 500-word instruction gets re-read every time. Let Claude ask you clarifying questions — you pick from options instead of writing paragraphs.

6. Batch Tasks Into One Message

3 separate prompts = 3 full context reloads. 1 prompt with 3 tasks = 1 reload.

7. Edit, Don't Reply

Click Edit on your message and regenerate. It replaces the old exchange instead of stacking follow-ups that grow the context.

8. Target Sections, Not Full Redos

Section 3 is wrong? Say "only redo section 3." Add: "No commentary. Just the output."

After the Chat

9. Summarize & Start Fresh Every 15–20 Messages

A 20-message session is roughly 105K tokens. A 30-message session is roughly 232K. Summarize, copy, new chat, paste.

10. New Topic = New Chat

Old messages are dead weight. Claude re-reads your first given task while you are onto the final stage. Start a fresh chat for each new topic.

11. Set Up Preferences & Styles

Personal Preferences + Custom Style eliminates 3 to 5 setup messages per chat. One setup, permanent savings.

12. Use Projects for Recurring Files

Upload once, cached forever. Every new chat references it without re-reading.

Quick Reference

| Phase | Action | Token Impact | |-------|--------|-------------| | Before | Convert PDF → .md | Saves 50–70% | | Before | Use Haiku for drafts | Saves 60–80% | | Before | Turn off Search/Connectors | Saves 10–30% per msg | | During | "Ask me questions" | Saves 80% vs long prompts | | During | Batch tasks | Saves 66% (3→1 reload) | | During | Edit, don't reply | Saves context growth | | After | Summarize at 15–20 msgs | Prevents 2x token bloat | | After | Projects for files | Eliminates re-upload cost |

Based on Ruben Hassid's "How to Stop Hitting Your Claude Limit" infographic from how-to-ai.guide