On 30 July OpenAI announced lower prices for GPT-5.6 Luna and Terra, and a new fast mode in the API. Their cheapest model gets five times cheaper. This isn't news about a model. This is news about the bill at the end of the month.
- GPT-5.6 Luna gets 80% cheaper, Terra 20%; the lower prices also apply to how usage counts toward paid subscriptions.
- A new Fast mode in the API replaces the old priority one: up to 2.5 times faster at double the price, with no change in capability.
- The performance numbers are OpenAI's own, measured on tests of their choosing - take them with a caveat.
A day earlier the same company was explaining how the new model came out cheaper to run. On 30 July came the follow-up: the savings pass down to the customer. It sounds like an ad. But the numbers are concrete - you can check them in the price list.
Who wins from this? First - those who already have volume. Run thousands of requests a day, and an 80 percent discount is the difference between an expense and a line item you can plan around. Second, and more interesting for us: the things that just weren't worth it before now win. Automation that saves twenty minutes a day was foolish at the old price. At the new one it makes sense. The ceiling of what's practical is now moved by price, not capability.
The caveat is mandatory. The comparisons are theirs, measured on their own tests, and among them is a claim that on a given test their model beats a rival's at almost a hundred percent lower cost per task. Numbers like that are advertising until an independent party repeats them. Separately: the cheaper model isn't the smarter model. Making the fast one cheaper often just means the heavy work stays with the expensive one.
Price is now a feature of the product, not a footnote at the end of the contract. You choose a model the way you choose a material - by what it carries, not by which is newest. Before you get excited about the discount, check which model actually does your job. A discount on something that doesn't do your job isn't a discount.