There is a difference between needing frontier capability because one is solving a truly open ended problem, and needing a reliable workhorse model to do something well understood. Local models (like Qwen 3.8 27B) have gotten so good that they can do all routine tasks at a fraction of the cost of frontier models.
This is open model intended for local usage. So, it doesn't matter. If the end user wants to run this on their own hardware, cost is defined by the electricity price + the price of the processing power. Providers can increase the price as long as the users don't switch for buying the hardware themselves instead.
