A training approach where a model learns to achieve specific target states or goals provided as input, rather than following fixed step-by-step instructions.
Multi-step reasoning, logic puzzles, mathematical problem-solving
Function calling, structured output, agent-style tool orchestration