GPT-5 was in no small part a "cost down update" for OpenAI - they replaced their default 4o with a more optimized, more lightweight option that they can serve at scale without burning a hole in their pockets. At the same time, their "top end" options for the power users willing to pay for good performance remain competitive.
The entire reason why OpenAI is burning money is "their investments to remain competitive". Inference is profitable - R&D is the money pit. OpenAI is putting money into more infra, more research and more training runs.