There are many 1 off use cases where LLM free versions look great, but when you invest to make a system of it (engineers/time/model servers), and then consider pricing, and realize most early adopters would rather use their custom contexts that they get for 1 off pricing (or free)