Skip to content
BSWEN
Blog

What Are Claude Code's Actual Usage Limits? (The Transparency Problem)

Cowrie
Cowrie
Dev @ Bswen

I was coding with Claude Code the other day when I hit a wall. The CLI showed “41% usage” and then immediately threw “Session limit reached.” Wait, what? I was only at 41%.

This made me wonder: what are Claude Code’s actual usage limits?

The Problem: Percentages of Nothing

I looked for documentation. Nothing. I checked the Anthropic help pages. Nothing useful. I asked customer support. They said there are “no active incidents related to token tracking.”

But here’s the thing: I don’t even know what my token cap is supposed to be.

Other users have the same confusion. On Reddit, someone posted that their usage showed 41% but they still got “limit reached” errors. Another user on the Max plan said it “feels like my max plan was downgraded back to Pro plan usage.”

How Other Services Handle This

Every other metered service I use is transparent about limits.

Cell phone plans tell you exactly how many gigabytes you get. Cloud providers show CPU hours and storage in gigabytes. Even other AI APIs like OpenAI show exact token counts.

How limits are typically shown
| Service         | What You See                    |
|-----------------|--------------------------------|
| Cell phone      | "15 GB per month"              |
| AWS             | "1000 CPU hours included"      |
| OpenAI API      | "Used 50,000 / 100,000 tokens" |
| Claude Code     | "41%" of ???                   |

The fact that Claude Code only shows a percentage with no denominator is… strange.

What I Tried

I tried to reverse-engineer the limits by tracking my own usage.

My attempt to measure
Session 1: 41% shown, limit hit after ~10 minutes
Session 2: Started fresh, hit 40% in 10 minutes with 2 parallel sessions
Session 3: Single session, lasted about 45 minutes before hitting limit

The numbers were inconsistent. Sometimes 41% would let me keep working. Sometimes I’d hit the limit at 30%. There was no pattern I could find.

Hidden Variables I Can’t Measure

The more I thought about it, the more I realized there are hidden factors affecting my usage that I can’t see:

Unknown factors affecting limits
- Which model am I using? (Opus vs Sonnet vs Haiku)
- How big is my context window?
- How many files did I read?
- How many tool calls happened?
- Thinking/reasoning tokens - are they counted?
- Did I process any images?

Any of these could explain why my “41%” varies so much. But I have no visibility into them.

What Users Actually Want

Reading through the Reddit thread, the ask is pretty simple.

User requests from the community
1. Show token caps clearly (e.g., "123,000 / 300,000 tokens used")
2. Real-time token counter during sessions
3. Predictable session durations we can plan around
4. Documentation explaining how percentages are calculated
5. Communication when limits change

One user put it perfectly: “I struggle to think of an industry where people were just told ‘usage limits’ and those limits were not actually quantified.”

The Trust Problem

The bigger issue is trust. Without quantified limits, Anthropic can change the rules whenever they want.

A Max user waking up to Pro-level limits has no way to prove it happened. They just have to accept it.

The transparency gap
What we know:    A percentage (e.g., 41%)
What we don't:   What that percentage is of
What we can't:   Compare plans objectively
What we fear:    Limits changing without notice

Summary

Claude Code’s usage limits are fundamentally non-transparent. Users pay for plans without knowing what they’re actually getting. Unlike every other metered service, there’s no quantified limit - just a percentage of an undefined total.

This opacity makes it impossible to:

  • Budget for projects
  • Compare plan value objectively
  • Predict session duration
  • Trust that limits won’t silently change

The solution is simple: show us the token counts. Give us the denominators. Let us make informed decisions about which plan fits our needs.

Until then, we’re all just guessing at what we’re paying for.

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: