o3-mini (high) uses 1/3rd of the compute of o1, and performs about 200 Elo higher than o1 on Codeforces.

o1 is the best code generation model according to Livebench.

So how is this not a breakthrough? It's a genuine movement of the frontier.