Anthropic introduces effort control for Claude API, allowing dynamic adjustment of model intensity mid-conversation

Anthropic has rolled out a new Beta feature to the Claude API, enabling developers to modulate the effort level during a single conversation thread. This functionality allows for dynamic shifts in the model's intensity, moving beyond a fixed setting for the entire dialogue. This effort setting functions as a dial controlling the computational resources dedicated to the response, balancing factors such as the depth of the answer, response speed, and operational cost. Adjusting this parameter influences more than just the final word count; it affects the resources allocated during the model's reasoning, tool invocation, and response structuring. Technically, users can implement this by inserting a system message with no content into the message sequence and setting the desired level via the new output_config.effort parameter. Crucially, this modification does not alter previously generated content and only takes effect starting from the next user message containing new input. This allows developers to seamlessly transition a conversation from simple Q&A to complex analysis by changing the effort level at the point of task switching, without needing to modify top-level configurations.