Anthropic says Claude’s lower performance is due to reduced Effort setting and launches Reflect usage dashboard

In early March, developers reported that Claude Code appeared to perform worse, skipping files, not running tests and asking for more input. Anthropic explained that on 4 March the default Effort level was changed from High to Medium to lower response latency. The company clarified that Effort determines how much computational work the model invests, not its underlying intelligence, so a lower setting can make the AI read less code and conduct fewer checks.

Anthropic also released Reflect, a beta analytics feature for Claude that compiles usage statistics such as conversation count, peak hours, frequent topics and time spent. Reflect offers tips like using Projects for recurring contexts and setting quiet‑hours reminders. The data used for reports comes only from eligible conversations, excludes private or health‑related content, and is not repurposed beyond the user’s dashboard. The feature is available on the web and desktop Claude apps.