From the course: Building with the Claude API by Anthropic

Unlock this course with a free trial

Join today to access over 25,500 courses taught by industry experts.

Extended thinking

Extended thinking - Claude Tutorial

From the course: Building with the Claude API by Anthropic

Start my 1-month free trial Buy for my team

Extended thinking

“

Let's examine one of Clawd's more advanced features, known as Extended Thinking. Extended Thinking gives Clawd time to reason about the user's query before generating a final response. In many chat UIs, this will be displayed as a separate thinking process, which the user can optionally look at to get better idea of how Clawd is approaching their problem. Now, in general, enabling Extended Thinking will allow Clawd to tackle more complex tasks with greater accuracy, but there are some big trade-offs here. You are charged for tokens generated by Claude during the thinking phase, and the phase itself takes some amount of time to complete. So with increased intelligence comes increased cost, and there's increased latency as well. Now a common issue around extended thinking is deciding when to enable it. And the answer to this is really simple. You're going to rely on your prompt evals. So you're going to write out a prompt, you're going to run an eval on it, and if the accuracy is not…

Contents