Structured Outputs with LLMs: JSON Mode, Function Calling, and When to Use Each

Getting structured, machine-readable outputs from Large Language Models is crucial for applications that need to programmatically process data. Two primary methods for this are JSON Mode and Function Calling. JSON Mode is a simpler approach that instructs an LLM to always return a valid JSON object, but it does not guarantee a specific schema or structure. In contrast, Function Calling is a more advanced technique that allows developers to define a specific schema, ensuring the model's output consistently matches the required fields, types, and structure, which is also foundational for building agentic AI workflows.