3
“While prompting generally works the same with extended thinking enabled, we’ve made it easy to optimize prompts to make the most of this new capability,” the company wrote, adding that users would need to specify which prompts should work with extended thinking.
In addition, Anthropic is also offering a budgeting capability for extended thinking inside the Console, which can be used to set a limit to the maximum amount of “thinking” tokens being generated.
The budgeting capability, according to Gupta, will allow enterprises to become better at problem-solving without the risk of overspending. “With this feature, enterprises gain control over when they want their models to think and by how much,” Gupta said.
