GPT-6 Astra reportedly uses loop transformers — more depth, not more parameters

zephyr_z9 · x · 2026-09-15

Per SemiAnalysis, GPT-6 Astra is "basically confirmed" to use loop transformers: instead of adding parameters, the model passes through its layers multiple times, buying compute depth without growing model size. The takeaway: labs are best positioned to know which way scaling works, and this is a tell that parameter count is no longer scaling aggressively in their roadmaps.

Original post →

More from Models

Models channel →