OpenAI cut GPT-5.6 Luna's API price by eighty percent and Terra's by twenty percent on 30 July, and replaced Priority Processing with a mode called Fast.
An eighty percent cut is not a discount. It is a repricing of what a token is worth, and it changes which products are viable to build.
What eighty percent actually unlocks
At the old price, a feature that ran a model over every row of a table, every incoming email or every page of a document had to justify itself line by line. At one fifth the cost, the same feature stops needing a business case.
This is why Luna got the deeper cut. Luna is the cheap tier — the one used for classification, extraction, routing and bulk work, where volume is high and per-call quality demands are modest. Cutting the cheap tier hardest is a bet on volume.
Priority Processing becomes Fast
The second change is quieter and worth reading carefully. Priority Processing was a paid guarantee of faster service. Fast mode replaces it.
If you have a production path that depends on Priority Processing behaviour, this is a migration, not a rename. Check your latency assumptions before the old path stops behaving the way your code expects.
The wider pricing picture
July's pricing news reads in one direction. Cursor shipped Router on 22 July to route requests to cheaper models, then a ₹649 India plan on 28 July, and OpenAI cut prices on 30 July.
For anyone paying for AI usage, the practical advice is unglamorous: re-check your bill assumptions quarterly. A number that was true in June was wrong by August.