ByteDance Puts Rival Models Inside Its Own Agent: Doubao Work Adds Kimi K3.1 Pro, GLM-5.3 and DeepSeek for Enterprise Users

ByteDance has made a quiet but strategically loaded change to its enterprise agent product. Doubao Work, the company's AI office agent for business customers, opened to third-party models on October 29, and the first batch is a who's who of its domestic competitors: Kimi K3.1 Pro from Moonshot AI, GLM-5.3 from Zhipu and DeepSeek-V4.1-Flash, all of which now appear in the app's model picker next to ByteDance's own Doubao models.

The mechanics are deliberately friction-free. The update applies to team and enterprise subscription plans and is switched on by default. Companies do not need to buy access from each model provider or file a separate application; usage is deducted from the AI credits the organization already holds in Doubao Work. Administrators control who can use which model, granting access to individual members, departments, groups or the whole company, and can enable every model from one provider at once or a single model on its own.

The move changes what Doubao Work is selling. Until now the app was largely a showcase for ByteDance's own models inside an agent that works across files, browsers and desktop apps. Letting enterprises route tasks to rival Chinese models makes the agent layer, rather than the underlying model, the main product. It also lets teams choose models by task, pairing a coding-focused model for engineering work with a different one for long-document analysis.

There is a pricing logic underneath as well. OpenRouter listings show GLM-5.3 at about $0.0699 per million input tokens, roughly a tenth of DeepSeek-Pro's $0.1035 prompt tier, and Doubao Work's credit system means enterprises can arbitrage those differences without new procurement. ByteDance, which operates one of the largest inference fleets in China, can afford to be a neutral marketplace for model access because it collects the agent relationship either way.

The pattern is not unique to ByteDance. Chinese desktop agent products have been converging on the same posture, and Tencent's WorkBuddy, which passed 32 million monthly users this week, has pursued a similar cross-app agent strategy with its own ecosystem integration. What distinguishes Doubao Work's version is the direct inclusion of competitors' frontier models in the default picker, an acknowledgment that in the Chinese enterprise market no single lab, including ByteDance's own Seed team, wins every task.

For the model vendors, the trade-off is exposure versus dependence. Kimi, GLM and DeepSeek gain distribution inside an office product with a large installed base and no integration work, but they also cede the customer relationship and the billing interface to a rival platform. That mirrors the dynamic familiar from Western marketplaces, where the distributor captures the margin and the model becomes a metered utility.

The signal for the broader market is that the agent layer is where Chinese AI competition is settling. As open and closed models converge on similar benchmark scores, the durable position is the product that orchestrates them, holds the enterprise credits and remembers the organization's context. ByteDance has decided it would rather host that fight than win every model benchmark.