Dev burns 300M tokens on GLM 5.3 in a week and still has quota left

saibharadwaj · x · 2026-09-10

Developer saibharadwaj notes that Chinese models like GLM, DeepSeek, StepFun and MiMo are surprisingly fast, offering huge token limits and bonus tokens with no resets — he consumed about 300M tokens on GLM 5.3 in a week and still had 2% left. He also wonders how Chinese AI firms keep building powerful custom AI chips despite semiconductor export restrictions.

Original post →

More from Infra

Infra channel →