“The o1 system is inefficient for simple queries, using approximately 10 times more tokens than a standard autoregressive model for the same task.”