If the next best model is 0.5x as expensive, then many might opt for that in use cases where the results are already good enough.

At work we are optimising cost by switching in different models for different agents based on use case, and where testing has demonstrated a particular model's output is sufficient.