When Will Open-Source Models Hit 3T Parameters?

TheZachMueller · x · 2026-07-15

A discussion around the feasibility of open-source models reaching the 3T parameter scale. A cited tweet suggests that it only takes one more iteration of open-source SOTA (like a potential K3 model) and a whitepaper on distillation to drastically shift the industry landscape. It notes that paired with good UX, current open-source models can already meet 90% of the needs for 90% of users.

In the replies, to the question of how to "sanely" run a 3T parameter model, someone provided a direct hardware answer: 8x8xB200 (meaning 8 nodes, each equipped with 8 B200 GPUs).

Related event: Open-weight models already 'good enough'? A 3T base and a distillation whitepaper seen as the watershed(5 posts)→

Original post →

More from Infra

Infra channel →