Model distribution formula and LLM cost calculation with provider-specific cached token accounting.
