Anthropic adds effort control to Claude API for mid-chat adjustments

Anthropic has added a new Beta feature to the Claude API. Developers can now adjust the effort level in the middle of a conversation. Effort controls how much work the model puts into a reply, balancing answer quality, speed, and cost. It affects thinking, tool use, and how answers are organized. To use it, developers just insert an empty system message into the chat and set the target level with the new output_config.effort option. No top-level changes are needed, and prompt caching still works. The new setting only applies from the next user message and doesn't change past replies. For example, if a chat starts with simple questions and later moves to complex analysis, developers can raise the effort level at the switch point without changing the overall setup, balancing quality and cost more easily.