Rumor: Grok Has 1.5T Parameters and Trend of Model Intelligence Compression

adonis_singh · x · 2026-07-12

Tweets discuss rumors and trends in current large model parameter scales: since GPT-4, model parameters have remained at 1-5T (trillion) level, and are now gradually moving toward 10T (rumored Fable and GPT-6). The author points out a clear phenomenon of "intelligence compression", e.g., rumors say Grok-4.5 has fewer parameters than GPT-4.

Related event: LLMs Approach 10T Parameters Amid Intelligence Compression Trend(4 posts)→

Original post →

More from Models

Models channel →