Severe latency regression on Opus 4.7 — ~20 min for simple prompts (was ~1 min 2 months ago)

Summary

Claude Code has become dramatically slower over the past ~2 months. Prompts that previously completed in ~1 minute now take ~20 minutes of thinking/response time, even for trivial edits on small files. This is blocking my daily use.

Environment

• Claude Code CLI on macOS (Darwin 25.2.0, zsh)
• Model: Opus 4.7 (claude-opus-4-7)
• Plan: Claude Max

Timeline

• ~2 months ago (around March 2026): reasonable latency, ~1 minute per simple turn.
• Progressively worse since.
• Today: ~20 minutes wall-clock for a single simple-edit turn on small files. Roughly a 20× regression for equivalent work.

Repro

1. Open a small project (few files, short files).
2. Ask for a trivial change — e.g. "rename this function", "add a prop", "fix this typo".
3. Observe the thinking/response time before any tool call lands.

Expected: sub-minute response on trivial edits, matching behavior from ~2 months ago.
Actual: ~20 minutes per turn.

/fast mode made it worse, not better

I tried /fast mode to work around the slowness, and on the same small prompt against the same small files it burned through roughly $20 or more of Max-plan usage for a single task. So the "workaround" trades latency for cost without actually delivering speed — net regression in both dimensions.

Asks

• Is this a known regression on Opus 4.7?
• Is there a server-side issue with routing/thinking budget for Max users on Opus 4.7?
• What's the recommended workaround? Should Max users default to Sonnet 4.6 or Haiku 4.5 for these tasks now?
• Happy to attach session telemetry / a transcript if you point me at where to send it.

Impact

20× slower + /fast burning through plan credits = Claude Code is currently not usable for the iterative small-edit workflow it used to be great at.