How are SaaS founders handling rising LLM costs?

Curious what everyone's approach is. Are you: Passing costs to customers? Limiting usage? Switching models? Using some kind of routing or optimization setup? At what point did AI inference costs become something you actively had to manage?