On September 18, two of China's leading AI labs launched new models on the same day, highlighting the intensity of the domestic model race on both raw speed and full-modality capability.
Zhipu officially released GLM-5.3-FlashX with its API fully opened. The new version delivers inference speeds up to 200 tokens per second — about five times faster than the existing GLM-5.3-Flash — while its pricing rises to 2.5 times the previous version, reflecting a premium on throughput.
The same day, Alibaba's Qwen launched a new-generation native full-modal model, Qwen3.8-Omni-Flash. Supporting text, image, audio and video input with a 1-million-token context, its core aim is to move full-modal models beyond 'understanding content' toward 'planning tasks, calling tools and completing creation.'
The burst of releases came against a policy backdrop: Wuhan recently issued its 15th Five-Year digital-intelligent economy plan, targeting an AI industry scale of 200 billion yuan by 2030 and 300 excellent agent products. On the capital markets, AI-themed ETFs strengthened, with one AI index up more than 2% intraday and related ETFs drawing over 70 million yuan of net inflows in five sessions.
Analysts at Tianfeng Securities said AI budgets in manufacturing, government, finance, healthcare, education and transport are likely to be released faster, accelerating tenders for agent platforms, industry software, data governance and compute services — pushing agents from pilot projects toward scaled deployment.
