Anthropic Clarifies Claude's Performance Drop: Not a Weaker Model, but Lower Effort Default

Anthropic responded to user complaints about Claude Code's perceived decline. On March 4, they lowered the Effort default from high to medium to reduce latency, reducing the model's investment in reading files, running tests, and multi-step tasks. Many developers mistook this for model degradation. AMD AI head Stella Laurenzo found Claude's thinking had dropped 67% via 6,852 session logs. On April 7, Anthropic restored the high default and reset usage quotas. Anthropic's post distinguishes Model (capability ceiling, "brain") from Effort (computational investment, "attitude"). A small model with high Effort can outperform larger models with low Effort, consuming up to 7x more tokens for verification. Claude Code now also offers an 'ultracode' tier for spawning multi-agent parallelism on heavy tasks. For engineers using Claude Code, this provides a decision framework: first check context and tools completeness, then determine if the issue is capability (upgrade model) or effort (increase Effort). Blindly switching to larger models wastes tokens and may not yield better results if Effort is low.