Alibaba has put the models of its two strongest Chinese rivals inside its own office agent. Qianwen Office now offers DeepSeek V4 Pro and Zhipu's GLM-5.3 as one-click choices in a "frontier models" tier, per a report dated 16 August Beijing time.

What is on the menu

DeepSeek V4 Pro is listed with a 1,000,000-token context and up to 384K of output, pitched at chaining long agent tasks. GLM-5.3 is presented as Zhipu's newest open-source flagship, claiming a 50% improvement over GLM-5.2 from further post-training, and aimed at long-horizon engineering and white-box code review and vulnerability discovery.

The speed is the signal

GLM-5.3 was released on Saturday morning Beijing time. It was a selectable option inside a competitor's shipping product by Sunday lunchtime — roughly a day. That propagation speed is only possible because these are open-weight releases, and it is what turns Chinese frontier models into interchangeable components rather than destinations.

What Qianwen Office is

The product is Alibaba's enterprise agent, assembled by merging QoderWork, MuleRun and Wukong, and billed as the first to support desktop agent, cloud agent and enterprise collaborative agent simultaneously. Pricing: free at ¥0, individual standard ¥78 a month (from ¥98), premium ¥158 (from ¥198), enterprise standard ¥198 per seat per month, flagship on enquiry.

What this is not

It is not Alibaba retreating from Qwen, and it is not a partnership. No revenue share, licensing or investment terms are disclosed, and "first to launch" is a distribution claim, not exclusivity. What it is is Alibaba's application layer becoming model-neutral while its model layer keeps competing — the posture Microsoft took with Copilot, arrived at from the opposite direction.