GPT-6 Astra estimated at 4.2T@120B, with around 112 layers
A user estimates that GPT-6 Astra loops half its layers once. They say their simulation most often landed at 4.5–4.6T, but they chose a lower central estimate because they tend to overestimate model sizes.
TLDR
A user puts their central GPT-6 Astra estimate at 4.2T@120B and around 112 layers, with 50% of its layers looped once. They say their simulation most often returned 4.5–4.6T, but they adjusted lower because they tend to overestimate model sizes. They also say they have evidence that GPT-6 Sol and Luna, and Opus 5.5, are not looped language models.
Combined views
8.9K
5 Sources, first seen 5h ago