Anthropic Said Claude Got Dumber. Here’s What Actually Happened.

Member-only story

Last month, Anthropic quietly shipped three changes to Claude Code’s operating layer. The developer community lost its mind. Anthropic’s April 23 postmortem named the culprits: a reasoning effort downgrade on March 4, a caching bug that wiped Claude’s own thinking mid-session, and a 25-word verbosity limit on system prompt responses added April 16. The thread on Hacker News ran to hundreds of comments. Two camps formed within 48 hours.

Team Shrinkflation said the evidence was overwhelming. Stella Laurenzo, a Senior Director at AMD’s AI group, published an exhaustive audit of 6,852 Claude Code session files and 234,000 tool calls on GitHub showing a measurable collapse in reasoning depth. BridgeMind’s benchmarks showed Claude Opus 4.6’s accuracy dropping from 83.3% to 68.3%. Users on Reddit and X called it “edit-first laziness.” The phrase “AI shrinkflation” went viral. Accusations flew that Anthropic was silently throttling performance to manage compute costs while telling users nothing was wrong.

Team Nuance said slow down. The API and underlying model weights were never touched. Three distinct product-layer changes — not a single deliberate nerf — created…