Mystery solved: Anthropic reveals changes to Claude's harnesses and operating instructions likely caused degradation

Anthropic said recent Claude quality complaints were caused by product-layer changes, not by degraded model weights.

Three issues were cited: lower default reasoning effort, a caching bug that wiped thinking history on later turns, and stricter prompt limits that made responses less verbose.

The company said its API and core model weights were unaffected, and that the affected settings have now been reverted or fixed.

The post-mortem helps explain the widely discussed performance drop and shows how non-model changes can materially alter perceived AI quality.