[BUG] Claude Code Max x20 limits are too low compared to Codex Pro x20, and reliability has degraded

Resolved 💬 2 comments Opened May 25, 2026 by pavelbe Closed Jun 25, 2026

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?

I use both Claude Code and Codex for real development work, and right now the difference is becoming hard to ignore.

I’m on Claude Max x20 and also use Codex Pro x20. In practice, Codex gives me significantly more usable quota/headroom than Claude Code Max x20. With Claude Code, I keep hitting limits and getting blocked even though my usage is not extreme: one project, one terminal, normal daytime work only — not 24/7 coding.

The bigger issue is trust. Claude Opus used to feel much more reliable, but lately it feels degraded. I no longer trust it the way I used to.

I’ve had Claude Code damage my workflow multiple times: reverting git changes unexpectedly, deleting a folder by mistake, and making other destructive changes that forced me to constantly monitor it. Now, whenever I give Claude Code any serious task, I feel like I have to watch every step.

Codex has not given me the same kind of problems. In my experience, it is much more predictable, and it has not damaged my work or made the same kind of careless destructive mistakes.

For the highest Claude Code plan, this is not acceptable. If users are paying for Max x20, the product should be usable for serious daily development without constant lockouts and without needing to babysit every action.

Please increase the Claude Code Max x20 limits and improve reliability/safety around file and git operations. Otherwise, more developers will keep moving to Codex and other alternatives.

What Should Happen?

Claude Code Max x20 should be usable for normal daily development without frequent multi-day lockouts.
With one project, one terminal, and regular daytime usage, the highest-tier plan should provide enough quota to work reliably throughout the week. Quota consumption should be predictable and transparent.
Claude Code should also be safe around file and git operations. It should not unexpectedly revert changes, delete folders, or make destructive edits without explicit confirmation.

The expected behavior is that Max x20 works as a serious daily development tool, not something users have to constantly babysit or stop using for several days because the quota runs out too quickly.

Error Messages/Logs

Steps to Reproduce

Use Claude Code Max x20 for normal development work during a regular work week.
Keep the workload moderate and consistent: one project, one terminal, daytime work only, no 24/7 usage, no heavy parallel automation.
Use Claude Code for typical coding tasks: implementing features, refactoring, debugging, writing tests, and making file/git changes.
Track how quickly the usage limit is consumed and how often the account gets blocked.
Run the same kind of workload with Codex Pro x20 under similar conditions.
Compare usable working time, quota drain, lockouts, and reliability of file/git operations.

Actual result:
Claude Code Max x20 runs out too quickly and can block me for several days, even with one project, one terminal, and normal daytime usage. I also have to constantly supervise it because it has made destructive changes before, such as reverting git changes or deleting files/folders.

Expected result:
The highest Claude Code plan should support a normal developer working on one project during regular work hours without frequent multi-day lockouts. It should also be reliable enough around file and git operations that users do not need to babysit every action.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.150

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Warp

Additional Information

_No response_

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗