Zhipu Releases GLM-5.3-FlashX Model with Up to 200 Tokens/s

On September 18, Zhipu AI officially launched the GLM-5.3-FlashX model, with its API now available. The model supports generation speeds of up to 200 tokens per second and runs on inference compute powered by 100,000 domestically produced chips.

According to Zhipu, the GLM-5.3-Flash was previously introduced to global developers under the name “Ox Alpha,” gaining wide recognition and steadily increasing call volumes. To meet fast-growing demand, the company increased investment in infrastructure and optimized inference on top of its existing compute resources. Zhipu said the GLM-5.3-Flash has maintained strong intelligence among same-size models, and the latest speed upgrade enhances its overall competitiveness in intelligence, pricing, and speed.

Developers can now access GLM-5.3-FlashX through Zhipu’s open platform using the model key “GLM-5.3-FlashX”.