Press Enter to search · ESC to close

AI × Crypto

China’s Office AI Agents Become Biggest Customers of Open-Source Models

Tencent's WorkBuddy has become the largest customer for open-weight AI models in China, as major internet platforms converge on multi-model strategies for office agents. The shift creates new opportunities for model providers but also concentrates bargaining power, with scale and execution efficiency emerging as key competitive advantages.

China’s Office AI Agents Become Biggest Customers of Open-Source Models

Tencent’s WorkBuddy has emerged as the largest customer for nearly all open-weight AI models in China, speaking on the company’s third-quarter earnings call. The disclosure highlights a broader shift: major internet platforms are now aggregating multiple third-party models into their office productivity agents, transforming them into super-buyers for upstream AI model providers.

The move marks a strategic pivot. WorkBuddy, launched in March, initially supported third-party models from DeepSeek, GLM, and Kimi, but rivals ByteDance’s Doubao Work and Alibaba’s Qwen Office stuck with in-house models. That changed in late October when Doubao Work opened to third-party models, followed by Qwen Office in mid-November, which added 10 models including Kimi, GLM, DeepSeek, MiniMax, and Step. The convergence suggests a new competitive dynamic: access to diverse models is becoming table stakes, while cost control and execution efficiency emerge as differentiators.

Market Implications

For AI model developers such as Zhipu, Moonshot AI, and MiniMax, the rise of office agents with hundreds of millions of daily active users represents a massive new distribution channel. MiniMax disclosed that its anonymous model Space Bunny processed over 200 trillion tokens while being tested on WorkBuddy before its official release. Such volumes can significantly boost revenue for model providers, but also concentrate bargaining power among a few large platforms.

Large-scale procurement is already leading to discounted API pricing. Companies with higher call volumes can negotiate lower unit costs, creating a positive feedback loop: lower costs enable more competitive product pricing, which attracts more users and further increases volume. This dynamic could squeeze smaller model providers and intensify competition among platforms.

Beyond procurement, the total cost of completing tasks depends on the agent’s orchestration framework, or Harness. Research from ByteDance’s Seed team shows that different Harness designs can vary token consumption by nearly four times for the same underlying model and task. As a result, platforms that optimize model routing and execution efficiency can achieve significantly lower operational costs, even with identical model access.

For investors, the development signals a maturing AI ecosystem where value is shifting from model creation to efficient deployment and integration. Companies that can aggregate demand, negotiate favorable terms, and optimize execution will likely capture disproportionate margins. Meanwhile, model providers with strong performance and cost-efficiency stand to benefit from increased token volumes, but face pricing pressure from large customers.

Key Takeaways

  • Tencent’s WorkBuddy is now the largest customer for open-weight models in China, with rivals ByteDance and Alibaba following suit.
  • Office agents are becoming critical distribution channels for AI model providers, driving massive token consumption.
  • Scale advantages allow large platforms to negotiate lower API prices, creating a competitive moat.
  • Execution efficiency and Harness optimization are emerging as key cost differentiators.
  • Investors should watch for margin shifts from model developers to platform aggregators.

View original

Share
Risk notice This site provides news and information on the crypto, blockchain and Web3 industry for reference only and does not constitute investment advice or any promise of returns. Virtual currency-related activities are illegal financial activities in mainland China; digital asset prices are highly volatile; use at your own risk. This site does not provide trading, token issuance or related referral services.

Related Reading

Latest News

TREE NEWS share card
Long-press image above → Save to Photos / Share
Pitch us Feedback