Right now, the understanding of what is acceptable or not for excessive Assistant usage is strictly based upon the tokens generated. This is, however, misleading in what it truly costs Kagi - DeepSeek generates a huge number of tokens when it thinks, but it's notably an extremely cheap LLM, whereas Claude's prices are substantially higher than the rest of the industry's. I'm happy to change my usage patterns based on whether or not I need the more expensive LLMs for my task, but right now I'm actually incentivized to spam the most expensive LLM offered. I just think there may be a better way.
Some categorization of LLMs based upon cost, and some unit value in our account usage that represents a combination of tokens and cost (I know showing raw cost is probably a little crass, so some intermediary unit can be calculated)? T3 chat shows their more expensive models separately:
