[BUG] Edit tool calls hang/timeout on v2.1.91, all models, Max 20x plan, Windows 11

Resolved 💬 6 comments Opened Apr 3, 2026 by jam-time Closed May 23, 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?

Here's the body that claude code told me to use:

Extreme slowness / near-unusable performance in Claude Code

Environment

  • OS: Windows 11 Home 10.0.26200
  • Claude Code version: 2.1.91 (latest)
  • Plan: Max 20x
  • Models affected: All — Opus 4.6 (1M), Opus 4.6, Sonnet 4.6

Problem

Claude Code has become nearly unusable. Token generation is extremely slow and tool calls frequently hang/timeout before completing.

Observed performance:

  • 3.6k tokens in 90 minutes (~0.7 tok/s)
  • 4.8k tokens in 109 minutes (~0.7 tok/s)
  • Simple 2-word responses taking 30+ seconds
  • Edit tool calls timing out repeatedly — user has to cancel them
  • Thought time is minimal (1s), so it's not "thinking" — it's generation/streaming that's slow

Not the cause

  • Not the API itself — claude.ai web interface with the same models on the same account works at normal speed
  • Not config/hooks — removed most hooks, no improvement
  • Not model-specific — same slowness on Opus 1M, Opus, and Sonnet
  • Not session-specific — persists across fresh restarts
  • Not version-specific — already on latest (2.1.91)
  • Same configuration worked fine for weeks — no config changes preceded the slowness

Expected behavior

Normal token generation (30-80 tok/s) and sub-second tool call responsiveness, as experienced previously with the same setup.

Additional context

The setup includes custom hooks (UserPromptSubmit), an MCP server, ~14 agent definitions, and 4 plugins. This exact configuration has been in use for weeks with no performance issues until recently. The fact that claude.ai works fine on the same account rules out account-level throttling.

What Should Happen?

I send a prompt and claude does stuff and responds instead of doing nothing for 2 hours.

Error Messages/Logs

Steps to Reproduce

type claude into terminal. hit enter. type write 'hello' to sample.txt. hit enter. wait 2 hours and consider codex.

<img width="729" height="77" alt="Image" src="https://github.com/user-attachments/assets/dc4b9393-dbf7-4277-8693-3405bad64830" />

Claude Model

Other

Is this a regression?

Yes, this worked in a previous version

Last Working Version

2.1.90

Claude Code Version

2.1.91

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

I've tried everything that I or claude could think of. Nothing worked. Claude said to put it here. It's entirely unusable out of nowhere.

View original on GitHub ↗

This issue has 6 comments on GitHub. Read the full discussion on GitHub ↗