Spitting out nearly 500 tokens per second! Zhipu GLM-5.3 high-speed version quietly makes its debut.

What happened: Zhipu quietly opened testing of GLM-5.3 HighSpeed in its ZCode AI coding tool. Access is random for ZCode subscribers, with no stated application channel. Key numbers: Over 490 tokens/s reported in testing. Why it matters: The reported speed is several times current domestic mainstream models, though it is a beta figure and final scale service speed is unconfirmed. Faster inference is becoming a competitive focus, with prior high-speed models aimed at cutting multi-round coding agent latency.

Zhipu's GLM-5.3 debuts at nearly 500 tokens per second.

Zhipu's GLM-5.3 debuts at nearly 500 tokens per second. — impact context not separately stored; see What changed.