Anthropic deploys Claude 3.7 Sonnet with dynamically adjustable reasoning budgets

The hybrid architecture allows developers to throttle chain-of-thought token allocation from instantaneous standard inference up to 128,000 reasoning tokens per prompt.