[jongjong]

Yep spot on. Price does not equate cost. Especially in our current economy where profit has been artificially made a non-factor. To know the cost, you'd have to look at hardware resource usage per query. Given that recent models have over a trillion parameters, you need a huge amount of memory and CPU to process a query to get the electrons to traverse all these thousands of billions of ANN nodes and/or weights.

Ultimately, it may turn out that dumber models may be more economically efficient than smarter models once you ignore the investment subsidy factor.

Maybe, given the current state of AI, the economically efficient situation is to have lots of dumb LLMs to solve small, well-defined problems and leave the really difficult problems to humans.

Current approach, looking at pricing is assuming another AI breakthrough is just around the corner.