Gary Marcus asks: is Astra a pure LLM or an undisclosed neurosymbolic hybrid?

GaryMarcus · x · 2026-09-04

Gary Marcus questions the architectural transparency of GPT-6 Astra: is it a pure LLM? Does it have a harness? Is it a hybrid model? With no public details on its architecture, he presumes it calls symbolic tools like other recent systems, making it likely neurosymbolic — a question that bears directly on whether the ARC-AGI scores reflect scaling or engineering.

Related event: Gary Marcus Questions GPT-6 Astra Architecture Transparency, Sparks Feud(3 posts)→

Original post →

More from Models

Models channel →