[BUG] Usage Limit Jumps from ~80% to 100% (with proof)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Summary

Since the latest update, Claude Code appears to consume the remaining usage quota almost instantly once the session reaches approximately 75-80%. Instead of gradually using the final 20-25% of the available budget, it immediately jumps to 100%, ending the session.

This effectively reduces the usable quota by around 20% compared to previous behavior.

Environment:

• Windows 11

Claude:

• Current Version: 2.1.201
• Previous Stable Version: 2.1.193
• Tested Versions: 2.1.193, 2.1.199, 2.1.201
• Model: Opus 8.4 (never used the other models)
• Reasoning: Medium
• Fable 5: Never used during testing
• Usage Context: Only Claude Code is used (no Claude Chat usage during sessions)
• Extensions/Tools: A browser extension was installed but not used (the browser was not open during testing)
• Other Tools: Figma MCP and Git only

What Changed:

Before this issue began, I could comfortably use nearly the entire session budget.

For example, I regularly asked Claude to:

• Build complete pages from Figma
• Implement complex UI
• Add animations
• Refactor code
• Fix bugs after implementation

Even after completing substantial work, there was still enough budget remaining (the final ~20%) to fix implementation issues, polish the output, or complete unfinished work.

Now, once the usage reaches roughly 75-80%, the next request immediately consumes the remaining budget and the session ends.

The result is that the practical usable limit has been reduced from approximately 100% of the session budget to only 75-80%.

Important Notes

This is not explained by:

• More complex prompts
• Larger implementations
• Increased tool usage
• Background processing
• Token-heavy requests

The workload is actually lighter than what I was successfully running a week ago.

The current pages are primarily composed of text and visual assets with relatively straightforward UI. Previously, I could implement significantly more complex pages and still have enough remaining budget for follow-up fixes.

The behavior changed abruptly rather than gradually.

Impact

This significantly reduces the practical value of each session.

Previously I could:

• Implement a page.
• Review the result.
• Fix bugs.
• Polish the implementation.

Now the session typically ends immediately after the initial implementation, leaving no remaining budget for iteration or corrections, even though similar or heavier workloads were handled successfully only a week earlier.

https://github.com/user-attachments/assets/c798893d-9d17-4d02-9cbe-fe09d2f6812e

What Should Happen?

Expected Behavior

Usage should increase progressively until the full session budget is consumed, allowing the remaining budget to be used for follow-up fixes, refinements, and small implementation requests.

Actual Behavior

Once usage reaches approximately 75-80%, the next request immediately consumes the remaining quota and ends the session, leaving the last 20-25% effectively unusable.

Error Messages/Logs

Steps to Reproduce

Troubleshooting Performed:

Updated to 2.1.199 where the issue first appeared.
Reverted back to 2.1.193.
Updated again to 2.1.201, where the issue still persists.
Confirmed that automatic updates were disabled and then re enabled and that is why i now am on the 2.1.201
Verified that the same issue occurs across all three tested versions: 2.1.193, 2.1.199, and 2.1.201.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.201

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

_No response_