The speaker expresses significant surprise and strong praise for the efficiency of a particular model, highlighting that it feels "way faster to use" and results in a "cheaper bill" than anticipated, emphasizing this discovery as a "big deal."
To illustrate this point, they compare three specific models: Fable 5, GPT 5.5, and Grok 4.5. All three models achieved "essentially identical" scores: Fable 5 scored 77, while GPT 5.5 and Grok 4.5 both scored 76.
However, a crucial distinction lies in the "cost to produce those scores" and "token usage per task." Fable 5 and GPT 5.5 incurred "enormous" costs in this regard. In stark contrast, Grok 4.5 achieved "essentially the same work done" and "the same quality" at a drastically lower cost, metaphorically described as "on the McDonald's salary." The speaker concludes by forcefully reiterating that this cost efficiency for comparable quality is "such a big deal."