I asked a historian:

Two postmortems, neither quite "admitted to nerfing":

Sept 2025, infra bugs: "A small percentage of Claude Sonnet 4 requests experienced degraded output quality" [0], alongside "We never reduce model quality due to demand, time of day, or server load." [1]

April 2026, Claude Code: default reasoning effort was lowered from high to medium, plus a caching bug and a verbosity prompt. Per Anthropic, "The models themselves didn't regress, and the Claude API was not affected." [2]

So users were right that quality dropped, but the confirmed causes were bugs and a product default, not deliberate model degradation.

[0] https://status.claude.com/incidents/72f99lh1cj2c

[1] https://anthropic.com/engineering/a-postmortem-of-three-rece...

[2] https://texxr.com/handle/claudedevs

source: https://claude.ai/share/4435bbcf-d6df-44a0-b1db-f08a11858bc2