Story perspectives

Zhipu AI’s GLM-5.3-Flash Hits 62 Trillion Tokens

8/28/2026

Summary

Story summary

Zhipu AI launched GLM-5.3-Flash, a model running entirely on 100,000 domestically produced AI chips.
During its stealth trial, GLM-5.3-Flash processed 62 trillion tokens, including 11 trillion in the first three days.
OpenRouter data placed GLM-5.3-Flash first among coding systems and top of weekly usage, surpassing DeepSeek.
Zhipu AI shares closed over 12% higher at HK$1,160, highlighting China’s domestic AI capability amid U.S. export restrictions.

History

Historical Context

China launched its AI-chip drive in 2008 when the Chinese Academy of Sciences funded the Cambricon project with 10 million yuan. After spinning out in 2016 and surviving U.S. sanctions, Cambricon supplied domestic chips that Zhipu AI used in 2024 to run its GLM-5.3-Flash model on 100,000 locally made processors, illustrating Beijing’s long-term push for semiconductor self-reliance.

1 of 2

Story summary

Zhipu AI launched GLM-5.3-Flash, a model running entirely on 100,000 domestically produced AI chips.
During its stealth trial, GLM-5.3-Flash processed 62 trillion tokens, including 11 trillion in the first three days.
OpenRouter data placed GLM-5.3-Flash first among coding systems and top of weekly usage, surpassing DeepSeek.
Zhipu AI shares closed over 12% higher at HK$1,160, highlighting China’s domestic AI capability amid U.S. export restrictions.

1 / 2