[BUG] Claude Code stuck in 1M context mode after model switch, blocking Pro user
Resolved 💬 3 comments Opened May 12, 2026 by Densist Closed May 15, 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?
Hi Anthropic Support,
I'm a Pro plan user and encountered a bug in Claude Code (VS Code extension) that forced me to clear my entire conversation, resulting in lost context and wasted usage.
What happened:
- I started a session using Opus with high effort mode
- I switched to Sonnet mid-session
- After switching, Claude Code appeared to inherit a 1M context window, which Pro users should not have access to
- I received the error: "API Error: Extra usage is required for 1M context"
- The compact option was also not showing, leaving me with no way to manage context
- The only fix was to clear the entire conversation, losing all context and wasting significant usage to get back to the same point
This appears related to GitHub issues #42616, #45390, and #45847.
I'd also appreciate a review of whether my usage was fairly consumed given the bug forced me to restart.
Thank you
What Should Happen?
Expected behavior:
- Switching models mid-session should not cause the session to inherit a 1M context window on a Pro plan
- The compact option should remain available as a fallback
- Users should not be forced to clear their conversation to recover from this state
Error Messages/Logs
Steps to Reproduce
Environment:
- Plan: Pro
- Interface: Claude Code (VS Code extension)
- Trigger: Switching from Opus (high effort) to Sonnet mid-session
Claude Model
Not sure / Multiple models
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
don't know
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
VS Code integrated terminal
Additional Information
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗