Claude 3.7 Sonnet Adds Extended Thinking Budget Controls for API Users

Anthropic updated Claude 3.7 Sonnet on its API to support configurable extended thinking budgets, letting developers specify maximum reasoning tokens between 1,024 and 100,000. This allows precise tradeoffs between response latency and reasoning depth depending on task complexity. The model also received improvements to tool use reliability and multi-step agentic task completion in benchmark evaluations.