Skip to content
BSWEN
Blog

Why Does Claude Hit Usage Limits So Fast? (Fix Guide)

Cowrie
Cowrie
Dev @ Bswen

Problem

When I opened Claude Code and ran /init, I got this message:

Claude Code output
Token usage: 5,000 tokens
Context window: 96% used

I had just started. Five minutes later, I hit the usage limit.

But when I checked the dashboard, it showed only 6% usage. So why did I get a rate limit error?

This isn’t just my experience. On Reddit, users report the same thing:

  • “Got that infamous limit after 2 interactions” (31 upvotes)
  • “I am at 21% after 17 minutes after reset and 4 messages” (3 upvotes)
  • “Just got my limits back. Started a task. Immediately 4% of MAX 20 5h limit used” (7 upvotes)

Environment

  • Claude Pro subscription ($20/month)
  • Claude Code CLI (latest version)
  • macOS Sequoia
  • Mid-size codebase (~50 files)

What happened?

I was setting up Claude Code for a new project. Here’s what I did:

Terminal session
$ claude
> /init

The /init command scanned my project structure, read key files, and generated a CLAUDE.md file. Standard setup.

But then I saw the usage percentage jump to 96%. I tried to continue working:

Error message
Claude usage limit reached.
Your limit will reset in 4 hours 32 minutes.

I was confused. I had barely started.

How to solve it?

The Real Cause: Three Independent Limits

I dug into the Anthropic documentation and found something important: Claude uses three separate rate-limiting systems, not one.

ConstraintWhat It MeasuresTypical Limit
RPMRequests per minuteVaries by tier
ITPMInput tokens per minute400K-4M tokens
OTPMOutput tokens per minuteVaries by model

The dashboard shows only an aggregate daily percentage. But 429 errors trigger when any single constraint is exceeded.

That’s why I saw “6% used” but still hit limits. The dashboard didn’t reflect all three constraints.

Immediate Workarounds

1. Use off-peak hours

Anthropic’s March 2026 promotion doubled limits outside 8 AM-2 PM ET. Schedule heavy work for early morning or evening.

2. Break up large requests

Instead of one massive query, use multiple smaller conversations. Each session starts fresh.

3. Clear conversation context

Start new chats instead of long threads. Long conversations accumulate token overhead.

4. For Claude Code users

Terminal commands
# Use compact mode to reduce overhead
claude --compact


# Close unused terminal sessions
# Each maintains its own context

Monitor Actual Token Usage

If you have API access, check real usage with Python:

check_usage.py
import anthropic


client = anthropic.Anthropic(api_key="your-api-key")


response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}]
)


usage = response.usage
print(f"Input tokens: {usage.input_tokens}")
print(f"Output tokens: {usage.output_tokens}")
print(f"Cache creation: {usage.cache_creation_input_tokens}")
print(f"Cache read: {usage.cache_read_input_tokens}")

This shows you exactly where tokens go. The dashboard can’t.

The reason

I think the key reason is agentic tools consume tokens differently than chat.

When I ran /init, Claude Code:

  • Scanned my entire project structure
  • Read multiple files simultaneously
  • Generated a CLAUDE.md with project context
  • Made multiple background API calls

Each operation counted toward my limits. But the dashboard simplified everything into one percentage.

Also, hidden system overhead adds up:

  • System prompt tokens (not visible)
  • Tool definitions and schemas
  • Context caching operations
  • Background API calls

These “invisible” tokens don’t show in the message count.

Summary

In this post, I showed why Claude’s usage limits seem “broken.” The key point is that three independent rate-limiting systems operate behind the scenes, and the dashboard doesn’t accurately reflect all of them.

For developers using Claude Code, the token consumption is even higher because agentic tools make multiple background calls.

What you can do:

  1. Use off-peak hours (outside 8 AM-2 PM ET)
  2. Break up large requests into smaller conversations
  3. Monitor actual token usage via API, not just dashboard
  4. Consider lighter models (Haiku) for simple tasks

Final Words + More Resources

My intention with this article was to help others share my knowledge and experience. If you want to contact me, you can contact by email: Email me

Here are also the most important links from this article along with some further resources that will help you in this scope:

Oh, and if you found these resources useful, don’t forget to support me by starring the repo on GitHub!

Comments

Tags: