Not arguing with your point about training efficiency, but the degree to which R1 is a technical breakthrough changes if they were calling an outside API to get

Not arguing with your point about training efficiency, but the degree to which R1 is a technical breakthrough changes if they were calling an outside API to get the answers, doesn't it? It seems like the difference between someone doing a better writeup of (say) Wiles's proof vs. proving Fermat's Last Theorem independently.