Zhipu AI Releases GLM-5.3-FlashX API with Speeds Up to 200 Tokens/s
Zhipu AI has officially launched the API for GLM-5.3-FlashX, an optimized model version delivering accelerated inference speeds of up to 200 tokens per second. The speed enhancement was achieved by optimizing Zhipu's infrastructure stack running across 100,000 domestic AI chips. Ultra-fast inference is crucial for real-time interactive applications, agentic coding, and high-throughput production workflows. By offering faster output generation on domestic hardware, Zhipu AI strengthens its market competitiveness across intelligence, pricing, and latency. GLM-5.3-Flash previously gained developer attention under the stealth codename 'Ox Alpha' on platforms like OpenRouter. Developers can now call the accelerated version via Zhipu's open platform using the model key `GLM-5.3-FlashX`.
## BACKGROUND
GLM (General Language Model) is the flagship series of large language models developed by Zhipu AI (also known as Z.ai). GLM-5.3-Flash introduced architectural changes such as a hybrid sparse and linear attention mechanism to significantly reduce long-context serving costs while maintaining high reasoning capability.