OpenAI Chart Leaks GPT-6 Astra as Researcher Questions Scaling Returns

A leaked OpenAI log-perplexity chart appears to reveal GPT-6 Astra and internal models. Pedro Domingos cited the data to argue that log-perplexity has only dropped to 0.75 while training costs have risen by orders of magnitude, highlighting diminishing returns from scaling.

2026-11-23 ~ 2026-11-23 · 2 related posts