GPT-5.6 Luna dropped 80% in price, Terra 20% — nine months after the model shipped
The 30 July repricing cut the cheap tier by four fifths and the expensive tier by a fifth. Asymmetric cuts are a positioning decision, not a cost pass-through.
If falling inference cost were the whole story, both tiers would move together. An 80 percent cut at the bottom and 20 percent at the top is a deliberate widening of the gap, and it is aimed at volume that would otherwise go to open weights or a cheaper competitor.
The economics underneath are worth stating plainly. A model already trained has a marginal cost that only falls, so price at the low end is a strategy variable rather than a constraint. What an 80 percent cut signals is that OpenAI would rather have the workload at a quarter of the revenue than lose it entirely.
That is the behaviour of a market where capability at the mid-tier has stopped discriminating between vendors. The competitive question has moved to cost per useful token, which is where open weights compete hardest.
OpenAI — GPT-5.6 August updates, deployment safety hub → · CNBC — OpenAI to publicly release GPT-5.6, rolls out conversational AI models →