reasoning_content: model reasoning or thinking content when exposed by the providercontent: final assistant response contentdelta: incremental streaming payload<think>...</think>. Your application should treat these fields as provider-specific metadata and render only the final answer unless your product intentionally exposes reasoning traces.