Max plan subscribers experience severe API slowdowns with no status visibility

Problem

As a Max plan subscriber, I frequently experience severe API slowdowns where Claude Code becomes nearly unusable — requests timeout after 90+ seconds. This happens unpredictably and there is no official status page or degraded-performance indicator from Anthropic.

Evidence

• Network latency is fine (TCP connect: ~35ms, TLS: ~90ms, total: ~260ms)
• API calls timeout at 90s — the bottleneck is entirely server-side
• Measured via MCP probe tool (claude-probe)

Impact

• Embedded systems development workflow is blocked during slowdowns
• No way to know if the issue is temporary (minutes) or extended (hours)
• No way to plan work around peak server load times

Feature requests

1. Server load transparency — expose a public status endpoint or header (e.g., x-server-load) so tools like Claude Code can adapt (e.g., switch to Haiku when Opus is overloaded)
2. Max plan prioritization — Max subscribers should have guaranteed throughput or priority queue during high-load periods
3. Automatic model fallback — when the requested model is overloaded, Claude Code could offer to fall back to a lighter model (Haiku) instead of timing out

Environment

• Claude Code CLI, Linux
• Max plan subscription
• Model: Claude Opus 4.6