Model cost analysis: 10T Astra could cost up to $100 per million tokens

xeophon · x · 2026-08-30

Comparing current large models, Nemotron (550B) costs $2.40/M, Qwen MoE (397B) costs $3.50/M, while GLM Flash (320B) is cheapest at $0.50/M. Based on GPT-5.5/5.6 estimated at 2-3T params and $20/M, a 10T parameter Astra model could cost between $66 and $100 per million tokens. The discussion advises against focusing solely on parameter count.

Related event: Debate: Do parameter counts determine frontier model pricing?(9 posts)→

Original post →

More from Models

Models channel →