[BUG] Extreme latency (60-96 seconds) for trivial queries on Max $200/month subscription

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?

Simple queries like "Reply only: OK" are taking 60-96 seconds to complete on a Max $200/month subscription. This extreme latency causes downstream services using Claude CLI to timeout, creating the appearance of "failures" when requests would eventually succeed.

This is distinct from the usage limit issues in #16157 and #9094 - requests ARE completing, but the latency is concerning.

What Should Happen?

A 2-token response to "Reply only: OK" should complete in 2-5 seconds, not 60-96 seconds.

Error Messages/Logs

## Evidence

### Direct CLI Testing (5 sequential requests)

Test 1: OK | Time: 62.807s
Test 2: OK | Time: 65.080s
Test 3: OK | Time: 62.367s
Test 4: OK | Time: 72.839s
Test 5: OK | Time: 96.381s


**Success rate: 100%**
**Average latency: 71.9 seconds for a 2-token response**

### CLAUDE.md Context Overhead Test
To isolate whether project context was the issue, I tested from a clean directory:

| Location | CLAUDE.md Present | Prompt | Response Time |
|----------|-------------------|--------|---------------|
| Project directory | Yes (~25KB) | "Reply: X" | **66 seconds** |
| /tmp (clean) | No | "Reply only with the letter X" | **28 seconds** |

**Findings:**
- Large CLAUDE.md files add ~38 seconds overhead
- **But even with NO context, 28 seconds for a 1-character response is not great**
- Base latency issue exists independent of context size

### Network Verification

$ ping -c 5 160.79.104.10
5 packets transmitted, 5 received, 0% packet loss
rtt min/avg/max/mdev = 4.417/5.809/7.022/1.030 ms


Network latency is **4-7ms**. The 28-96 second response times are NOT network-related.

### Downstream Impact
When using Claude CLI programmatically:
- Services timeout waiting for responses
- Errors appear as "All connection attempts failed" or "Server disconnected"
- Actual cause: latency exceeds timeout thresholds

Steps to Reproduce

1. Have a Max $200/month subscription
2. Run: time claude --print --model haiku "Reply only: OK"
3. Observe response time of 60+ seconds for trivial query
4. Repeat to confirm consistency

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.3

Platform

Other

Operating System

Ubuntu/Debian Linux

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

• Timeline: Started late December 2025, correlates with holiday period changes
• Pattern: Consistent 60-96s latency across multiple tests
• Not a limit issue: Requests succeed, just extremely slowly