Dylan PatelFounder and Chief Analyst, SemiAnalysis · The semiconductor-AI stack

GPT-6 Astra reportedly gets deeper without getting bigger. The labs already know what that means for

15 Sept 2026 · 1 min

Watch on YouTube

0:42

GPT-6 Astra uses Loop Transformers, which process layers multiple times to add computational depth without increasing parameters. AI labs are pushing compute-heavy strategies, signaling they don't expect to scale up model size much further.

Chapters

Loop Transformers is a new, is a relatively recent one. And it's basically confirmed that GPT-6 Astra uses Loop Transformers, which means that instead of like adding parameter count and input, you know, goes through the layers sort of more than once, right? So you add compute depth, but you don't increase the size of the model. And then the other point is that we're seeing the loudest cries for all-high are coming from the labs. And the labs are probably the people that are best positioned to, to, to say which way models are scaling. And I think that's also a tell that they're not seeing parameter sizes scaling as aggressively in their roadmaps, right? In, you know, what they find in their research.