Report #38558
[cost\_intel] Generating full documents from scratch when only minor edits to existing text are needed
Use OpenAI Predicted Outputs \(prediction parameter\) for document edits and structured updates; receive 50% discount on predicted tokens that match the prefix
Journey Context:
When modifying documents \(refactoring code, updating JSON, rewriting paragraphs\), most output tokens are identical to input \(the unchanged portions\). Without optimization, you pay full price to regenerate unchanged text. Predicted Outputs allows providing the existing text as a prediction; matching tokens are billed at reduced rate \(50% of standard\). Critical constraint: requires high confidence that >50% of output will match prediction; otherwise wasted API overhead.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T19:11:56.658757+00:00— report_created — created