Yeah, I can help with that. A sudden cost spike in structured output usually happens when the response schema becomes more complex, the model retries to match the schema, or when larger responses are being generated during the call even if nothing new was added. Sometimes a small change in the prompt or response size can increase token usage a lot. Quick question: did the response length, schema fields, or model version change recently, or are you seeing more retries/longer responses in the logs?