OpenAI Changes How Businesses Pay for Repeated AI Tasks
AI Wallet
OpenAI has introduced new pricing options that could reduce the cost of repetitive AI tasks. Its Decisions API, using GPT-6 Luna, charges $0.10 per million input tokens and $0.01 for cached input, with no separate output-token charge. This could make tasks such as sorting documents, checking conditions and routing customer requests cheaper. However, the final cost still depends on how much information is processed and the applicable pricing conditions.
Illustrative assumptions: normal input $0.10, cached input $0.01, standard (30-minute) cache write $0.125 and extended (12-hour) cache write $0.25 per million tokens. The first request writes the cache; later requests read it while it is still alive. Not every request automatically qualifies for caching — only the repeated, unchanged start of a prompt can be reused, and a cache hit is not guaranteed.
Rates are OpenAI's published short-context prices; prompts over 272K input tokens cost 2× and regional processing adds 10%. Output tokens are not included (the Decisions API has no output-token charge).
OpenAI has also introduced longer prompt caching, allowing supported applications to reuse information for up to 12 hours instead of relying only on shorter cache periods. This can be useful for businesses that repeatedly use the same instructions, documents or reference information throughout the day. However, keeping information cached for longer costs more at the beginning. The financial benefit depends on how often the information is reused and whether the application qualifies for cache discounts.