Claude Opus 5.5 is the most-discussed AI model on X right now — for the wrong reasons. Developers say Anthropic's coding flagship, released September 22, has started fumbling simple edits: repeating rejected solutions, ignoring files it just read, degrading back toward Opus 5. The complaints crested November 4–6. The strongest evidence comes from one power user's logs: 6,638 Opus 5.5 answers scored from local Claude Code transcripts. Before October 19, 5.2% needed correction. After October 19, 25.4% did — with the same setup. The counter-evidence is real too. LiveNerf, which runs the same frozen test panel on Opus 5.5 every day, reports no change across three 10-day windows after launch week. One confounder matters: Anthropic's own docs say Opus 5.5 defaults to medium effort — one level lower than Opus 5. Anthropic has not responded publicly to the complaints as of November 7. The catch: nobody has proven a regression — or ruled one out. The tracker data says the model is fine. The user logs say something changed on October 19. Until Anthropic answers, both readings stay on the table. Opus 5.5 still leads most November frontier rankings, costs $4 per million input tokens — 40% less than Opus 5 — and scored 66.4% on Terminal-Bench 4.0. The question is whether the model developers are actually getting is the one those benchmarks measured. BTC sits near $75,900, down about 0.1% on the day.