[BUG] Thinking Mode causes extreme response delays (5-30+ min) in WSL2 environment

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?

When Thinking Mode is enabled, Claude Code takes an extremely long time to respond — consistently between 5 and 30+ minutes — even for simple greetings or basic questions. The issue occurs in WSL2 (Ubuntu on Windows 11) and has been consistently reproducible for over a month across multiple terminal emulators (Windows Terminal, VS Code integrated terminal, Antigravity).

The Claude.ai desktop app works perfectly under the same network conditions, confirming this is specific to the Claude Code CLI running inside WSL2.

What Should Happen?

Claude Code should respond within a reasonable time (seconds) regardless of whether Thinking Mode is enabled, including for simple prompts. Thinking Mode should function reliably in WSL2 environments just as it does in other environments.

Error Messages/Logs

No explicit error messages are shown. The observed behaviour is:
- Spinner appears with a continuously incrementing timer
- "Thinking…" state persists for 5 to 30+ minutes with no output
- Token count does not progress during the hang
- One confirmed case: 10 minutes 42 seconds to respond to a web search question (✻ Cooked for 10m 42s)

Running `ps aux | grep claude` inside WSL2 shows the following flag being passed by the Antigravity terminal:


--max-thinking-tokens 31999


Multiple Claude Code instances are sometimes found running simultaneously in the background.

Steps to Reproduce

1. Run Claude Code inside WSL2 (any terminal emulator)
2. Ensure Thinking Mode is enabled (default configuration)
3. Send any prompt — including a simple greeting such as "Hello, how are you?"
4. Observe: Claude Code enters "Thinking…" state and does not respond for 5–30+ minutes
5. Cancel with Esc and resend the same prompt — it typically responds immediately

Claude Model

Sonnet (default)

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.87 (Claude Code)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

• Claude Code version: 2.1.87
• Operating system: Windows 11 with WSL2 (Ubuntu)
• Terminals tested: Windows Terminal, VS Code integrated terminal, Antigravity
• Model: Sonnet 4.6 (Default)
• Login method: Claude Pro Account
• Issue duration: Consistently reproducible for over 1 month

Workarounds tested — none fully resolve the issue:

• Disabled Gmail MCP server
• Performed full logout/login via claude auth logout / claude auth login
• Reinstalled Claude Code
• Rebooted PC
• Tested with and without VPN
• Disabling Thinking Mode eliminates the problem but removes important functionality

The issue does NOT occur in the Claude.ai desktop app under identical network conditions, suggesting a specific incompatibility between Thinking Mode, the --max-thinking-tokens flag, and WSL2's virtualized network layer (SSE streaming instability).