Why is my OpenAI bill so high? The 6 usual culprits

1. You're re-sending the same prompt at full price

If a large system prompt or document is included on every call and prompt caching isn't enabled, you pay full input price to reprocess identical tokens each time.

Cached input bills at a fraction of the standard rate. On a repetitive chat or RAG flow this alone is often a double-digit percentage of the bill.

2. Everything runs on the top model

The price gap between tiers is enormous — often more than 10x from the cheapest to the most expensive model. Routing simple classification and extraction to a small model while reserving the frontier model for genuinely hard reasoning is the single largest lever.

3. Non-urgent work runs synchronously

Nightly jobs, bulk classification, evals and report generation don't need a real-time answer. The Batch API is 50% cheaper on both input and output for anything that tolerates minutes-to-hours latency.

4. Unbounded output

Output tokens cost several times more than input. Leaving max_tokens unset and using verbose formats quietly inflates the most expensive token class.

5. Duplicate and retry calls

A meaningful share of production queries are near-duplicates — same document, same question, or a retry storm when a response fails to parse. Without semantic caching you pay to generate identical tokens twice.

6. Old, overpriced model snapshots

Deprecated model versions often cost several times their current-generation replacement for the same or better quality. Migrating is a config change, not a trade-off.