Office agents are now officially entering the multi-model aggregation competition.

On the evening of August 14, Qianwen Office announced the launch of the GLM-5.3 and DeepSeek V4 Pro models, allowing users to select and use them directly from the "Frontier Models" section on the product homepage.

With the earlier integration of Qwen3.8-Max, Qianwen Office's frontier models now cover three model providers: Alibaba, Z.AI, and DeepSeek.

Among them, DeepSeek V4 Pro features a 1 million token context window and a maximum output of 384K tokens, focusing on enhancing long-range tasks and agent task orchestration capabilities. GLM-5.3 further improves capabilities in programming and complex long-range engineering tasks.

This means that following Tencent WorkBuddy, ByteDance TRAE Work, and Baidu KuKu AI, Qianwen Office under Alibaba has also moved toward multi-model aggregation.

Compared to relying solely on proprietary models, the logic behind the aggregation route is that the strengths of different models can be better leveraged. This is because each model provider has different advantages in areas such as coding, complex reasoning, long text, search, response speed, and call cost, while a complex agent task often requires completing multiple steps in sequence.

However, the more models are integrated, the more another question arises: which model should be used for the task?

Currently, Qianwen Office primarily relies on users to actively select models. According to the product's division, tasks such as complex plans and automated workflows can choose "Frontier Models," while lightweight tasks like quick research and table analysis can use the lower-cost "Economy Models."

Taking it a step further, the agent could automatically handle model selection in the background, which is known as "routing."

In an ideal scenario, users would not need to understand the strengths and weaknesses of each model, but would only need to submit a task. The background would then dispatch based on the task's difficulty, model capabilities, and call costs: simple information extraction would be assigned to a faster, lower-cost model, while coding and complex reasoning would be assigned to a more capable model.

For individual users, this directly impacts task completion quality, wait time, and token consumption. In enterprise scenarios, it further translates to computing cost management.

Moreover, if the office agent accumulates a sufficiently large user base, this layer could generate new data value. Each real-world task leaves feedback on model performance: which model is more successful for which type of task, how long it takes, how many tokens are consumed, and where failures are more likely to occur. This feedback, provided it complies with privacy, authorization, and data governance requirements, can in turn be used to optimize routing strategies.

Currently, this competition is still in its early stages. The industry has not yet reached a consensus on when to switch models, or whether to schedule based on a single request or an entire task sequence.

However, with Qianwen Office joining the model aggregation camp, the office agent competition among Tencent, ByteDance, Alibaba, and Baidu is gradually aligning on the same track. As models continue to proliferate, the next competitive battleground may be who is better at "using models."