TIME: Making Reasoning Models 10x Cheaper by Thinking Only When Needed

TIME introduces dynamic reasoning allocation for LLMs, reducing inference costs by 90% while improving accuracy. Instead of forcing expensive thinking traces on every query, the model learns when reasoning is actually needed - making production deployment practical.