Zhipu has said nothing publicly, yet a model named GLM-5.3-Highspeed is already appearing for some paying users inside ZCode, the programming assistant the company sells. Pandaily reported on 13 October, citing QbitAI, which cited a user who noticed the new entry in version 3.15.1 of ZCode. That user’s screenshot showed roughly 490 tokens per second.
Access is a gray release to randomly chosen subscribers. A QbitAI reporter on the Lite plan could not switch to the model, and it is unclear whether the plan or the draw explains that.
At 490 tokens per second, 10,000 tokens would take about 20 seconds. Artificial Analysis clocked GLM-5.3-Flash at around 60 on the Z.ai API, while GLM-5.3-FlashX peaked near 200. Pi, a coding agent whose code is public, had already listed the model in its catalog, with room for 1 million tokens of context. Reddit users have asked whether it relates to Omen Alpha, a nameless model that showed up on OpenCode in September. Zhipu has confirmed none of it.
A coding agent making many calls in a row pays for every slow one. Until Zhipu publishes its own number, treat 490 as a lead to test, not a spec to plan around.
Reported by Pandaily on 13 October 2026, citing QbitAI.
