Jake Mintz(@jakemintz):My very basic mental model is that model performance comes from 3 vectors: compute, data, and training/model architecture. Google has historically had immense advantages in the first two, which has made it so confusing that they've been not been able to ship a frontier model. Congrats to Koray and team on an impressive looking model and maybe an even more impressive shipping cadence given Argon was just released weeks ago. Can't wait to get my hands on Carbon. And hopeful this is the beginning of a revitalized Deepmind, the whole world benefits from more competition at the top.

2026.10.28 21:09

My very basic mental model is that model performance comes from 3 vectors: compute, data, and training/model architecture. Google has historically had immense advantages in the first two, which has made it so confusing that they've been not been able to ship a frontier model. Congrats to Koray and team on an impressive looking model and maybe an even more impressive shipping cadence given Argon was just released weeks ago. Can't wait to get my hands on Carbon. And hopeful this is the beginning of a revitalized Deepmind, the whole world benefits from more competition at the top.