Opus 4.6 recurring outages making Claude Code unusable for Max subscribers (March 17-18)

Problem

Claude Code has been effectively unusable since March 17, 2026 due to recurring Opus 4.6 server-side degradation. Sessions hang for 10-15+ minutes on simple prompts. This is not a local issue — your own status page confirms 4 separate Opus 4.6 incidents in the last 24 hours alone.

Impact

• All Claude Code sessions hang indefinitely when Opus is degraded — no timeout, no fallback, no error message
• Multiple concurrent projects blocked across terminals
• This is the third wave of outages this month (March 2, March 11, March 17-18)
• Paying Max-tier pricing for a service that has been intermittently down for most of March

Expected Behavior

Claude Code should not hang silently when Opus is degraded. At minimum:

1. Timeout and retry — if an API request takes >60s, surface an error instead of hanging forever
2. Automatic fallback to Sonnet — when Opus is returning errors, gracefully degrade to Sonnet 4.6 instead of blocking the user
3. Status awareness — Claude Code should check status.anthropic.com and warn users when there's a known incident instead of appearing frozen
4. --fallback-model should work in interactive mode — currently it only works with --print

Reproduction

1. Run claude --dangerously-skip-permissions during any of the recent Opus 4.6 incidents
2. Type any prompt
3. Watch it hang on "Kneading..." / "Spinning..." / "Manifesting..." for 10-15+ minutes
4. No error message, no timeout, no fallback

Environment

• Claude Code v2.1.78
• macOS Darwin 25.3.0
• Claude Max plan (Opus 4.6)
• Multiple terminals/projects affected simultaneously

Status Page References

• March 17, 7:47 PM UTC — "Elevated errors on Claude Opus 4.6" (~4.7 hours)
• March 18, 6:41 AM UTC — "Elevated errors on Claude Opus 4.6" (~3 hours)
• March 18, 12:30 PM UTC — "Increased errors on Opus 4.6" (~1.5 hours)
• March 18, 3:16 PM UTC — "Elevated errors on Claude.ai" (~50 min, Claude Code login/logout affected)

Request

This needs to be treated as a P1. Silent hangs with no fallback mechanism is unacceptable for a paid CLI tool. Users should never have to manually diagnose whether Anthropic's servers are down by checking a status page — the tool should handle degradation gracefully.