Developers report Claude Opus 4.6 and Claude Code are performing worse, with accusations of 'AI shrinkflation' and potential throttling by Anthropic.

Users across GitHub, X, and Reddit are complaining that Claude's flagship coding models have become less capable, less reliable, and waste more tokens than weeks ago.

Reported issues include degraded sustained reasoning, premature task abandonment, increased hallucinations, and contradictions in responses.

Some users frame the issue as 'AI shrinkflation'—paying the same price for a weaker product—while others suspect Anthropic may be intentionally throttling performance during high-demand periods.

Anthropic employees have publicly denied the company deliberately degrades Claude's performance, leaving the claims largely unproven.