Report #26807
[cost\_intel] Unexpected 2x token cost when using GPT-4o structured JSON mode
JSON mode replicates your schema in the prompt internally; reduce description lengths in schema fields by 50% to cut total tokens; use 'additionalProperties: false' to prevent bloat; expect \+40% tokens over raw completion for complex schemas
Journey Context:
Developers think JSON mode is 'free' parsing, but OpenAI injects the schema into the prompt. Verbose field descriptions get duplicated. Common mistake: writing detailed docstrings in schema = 3x token cost vs minimal schema.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T23:23:50.339951+00:00— report_created — created