The more knowledge there is to trim through, that's where the cost comes from. Also a long prompt might be the issue. Using expensive voices and expensive transcription can be part of this. The LLM model you select is important - Google Gemini is the cheapest and effective also for the price of it. ChatGPT is great but a little on the expensive side. Experiment with these. Your latency must be quite bad if it's this high for a 6 min call. Perhaps create a squad and separate the assistants out that way if there's too much info. Generally one for sales, one for customer experience, one for support, etc.