[BUG] compact

Open 💬 8 comments Opened Jun 4, 2026 by SuperEarthling

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?

What's wrong

Two linked problems on the Claude Code desktop app (Max plan, Windows):

  1. Auto-compaction never triggers. Across long sessions I have never once

observed an auto-compaction pause. Context just keeps growing.

  1. **At/near the 200K limit, the client silently issues a 1M-context request

that fails with a credits error, freezing the session. I never selected a
1M model — the model picker shows standard
Sonnet 4.6** and the context
bar shows /200.0k. The 1M request is invoked automatically.

The error first appears as a misleading outer message
("Server is temporarily limiting requests / Too many requests right now —
try again in a moment"). Clicking View details reveals the real error:

API Error: Usage credits required for 1M context · turn on usage credits at
claude.ai/settings/usage, or use --model to switch to standard context

This is a feature-gate issue, NOT a usage-limit issue: my plan usage is far
below limits (5-hour: 12%, weekly all-models: 15%, Sonnet-only: 22%).

Steps to reproduce

  1. Select standard Sonnet 4.6 (200K context) in the desktop model picker.
  2. Work through a long session so context grows toward the 200K limit.
  3. Observe: no auto-compaction ever occurs; instead the session freezes and

View details shows "Usage credits required for 1M context".

Expected behavior

  • Auto-compaction should actually trigger near the limit, OR
  • Compaction / overflow handling should fall back to the standard 200K context

the conversation is configured for, without silently requiring 1M context
and usage credits.

What Should Happen?

Expected behavior

  • Auto-compaction should actually trigger near the limit, OR
  • Compaction / overflow handling should fall back to the standard 200K context

the conversation is configured for, without silently requiring 1M context
and usage credits.

Error Messages/Logs

Server is temporarily limiting requestsToo many requests right now — try again in a moment. API Error: Usage credits required for 1M context · turn on usage credits at claude.ai/settings/usage, or use --model to switch to standard context

Steps to Reproduce

Steps to reproduce

  1. Select standard Sonnet 4.6 (200K context) in the desktop model picker.
  2. Work through a long session so context grows toward the 200K limit.
  3. Observe: no auto-compaction ever occurs; instead the session freezes and

View details shows "Usage credits required for 1M context".

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

I didn't notice this problem.

Claude Code Version

2.1.150 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

<img width="1530" height="784" alt="Image" src="https://github.com/user-attachments/assets/d8e54a20-078b-467c-949b-dc87e1f78cba" />

<img width="1546" height="784" alt="Image" src="https://github.com/user-attachments/assets/4faab182-81f5-4302-b75c-0c818a19d588" />

<img width="955" height="698" alt="Image" src="https://github.com/user-attachments/assets/e34858ff-8485-4423-9cc2-edfcf0e564c5" />

View original on GitHub ↗

This issue has 8 comments on GitHub. Read the full discussion on GitHub ↗