I’m trying Claude’s extended thinking on tasks where it needs to inspect a few records, call a tool, then use the result to decide what to do next. The extra reasoning seems useful on some prompts, but I’m not sure whether it changes the way the model plans or sequences those calls.

When you’ve used Claude extended thinking with tools, have you seen it make fewer unnecessary calls, or does it mostly add latency before the same tool usage? I’m interested in practical differences for multi-step tasks, not just answers that need more explanation.