[BUG] **Title**: Claude Code 2.1.233 intermittently stalls ~3 minutes before responding on Windows (client-side, zero network connections during stall)
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?
## Summary
After upgrading from 2.1.227 to 2.1.233, requests intermittently take ~3 minutes before the response appears (and
often fail with "Connection dropped (ECONNRESET) Retrying in 1s attempt 8/10"). The same config and the exact same API
call are fast on 2.1.220 (~4.5s). Colleague on the same Windows + 2.1.233 + same config does NOT see this.
## Environment
- OS: Windows 11 Home (22621), x64
- Claude Code: 2.1.233, npm-global install (native binary)
- Custom endpoint: ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic (DeepSeek Anthropic-compatible API)
- Model override: deepseek-v4-flash[1M]
- No HTTP(S)_PROXY env, system proxy off, no proxy software running
## Steps to reproduce
Run repeatedly: claude -p "hi" (same settings.json every time).
- ~6 out of 7 runs: response appears after ~3 minutes (180–192s).
- ~1 out of 7 runs: response in ~1 second.
## Key diagnostics (already isolated)
- DeepSeek responds in 1.5–2.5s when the EXACT captured request (identical headers + 92KB body) is replayed via
curl/Node against api.deepseek.com/anthropic/v1/messages. Response is a complete, valid SSE stream including
message_stop.
- During the ~3-minute stall,
netstat -anoshows the claude process has ZERO TCP connections — it is NOT
connected to the API, NOT in SYN_SENT, NOT waiting on an established socket. The stall is local client processing, not
a network wait.
- The request sent by 2.1.233 is effectively identical to 2.1.220's (same model, thinking adaptive, effort high,
stream true, same anthropic-beta header; only version/session-id differ).
claude --versionstartup is fast (~1.7s) on both 2.1.220 and 2.1.233, so it's not startup.- 2.1.220 performs the same call in ~4.5s. Machine-specific: reproducible on this machine only.
## Expected behavior
Response time comparable to 2.1.220 (~seconds).
## Actual behavior
Intermittent ~3-minute client-side stall (process alive, no network I/O) before the response completes; during the
stall, connection drops surface as ECONNRESET retries.
What Should Happen?
other
Error Messages/Logs
Steps to Reproduce
otehr
Claude Model
None
Is this a regression?
No, this never worked
Last Working Version
_No response_
Claude Code Version
2.1.233
Platform
Other
Operating System
Windows
Terminal/Shell
PowerShell
Additional Information
_No response_