QwenWork Adds GLM-5.3 and DeepSeek V4 Pro, Bringing Three Chinese Flagship Models Into One AI Office Agent

Introduction

Alibaba’s AI office agent QwenWork (千问办公) has expanded its model lineup with two newly released Chinese flagship models: GLM-5.3 and DeepSeek V4 Pro.

According to the original AIBase report, QwenWork announced the update on the evening of August 14, 2026. Users can now select the two models directly from the product’s Frontier Models tier.

Together with Qwen3.8-Max, which was already available in QwenWork, the platform now brings flagship models from Alibaba’s Qwen team, Z.AI, and DeepSeek into the same office-agent product.

The update marks a clear move toward a multi-model strategy: instead of tying every task to one foundation model, QwenWork can offer users different frontier models inside a single agent workspace.

GLM-5.3 and DeepSeek V4 Pro Join QwenWork

Both new models arrived only shortly before the QwenWork integration.

DeepSeek V4 Pro was officially released on August 13, while GLM-5.3 followed on August 14. Both are positioned as the latest flagship models from their respective developers, but they emphasize somewhat different strengths.

DeepSeek V4 Pro

DeepSeek’s current production model is listed in its API documentation as DeepSeek-V4-Pro-0813, exposed through the deepseek-v4-pro model name.

The official specifications include:

• A 1 million-token context window
• Maximum output of up to 384K tokens
• Thinking and non-thinking modes
• Tool calling
• JSON output
• Responses API support
• An Anthropic-compatible API
• Context caching

DeepSeek says the V4 Pro release strengthens agent capabilities, making it particularly relevant for long-running tasks and workflows that require several tool or reasoning steps to be connected together.

For an office agent such as QwenWork, those characteristics matter because many real workplace tasks are not single-turn questions. Research, document production, data processing, web operations, and project workflows can involve large amounts of context and a long chain of actions.

GLM-5.3

Z.AI released GLM-5.3 on August 14 with a strong focus on coding, long-horizon engineering work, and cybersecurity capabilities.

One of the notable aspects of the release is that Z.AI describes the improvement as coming from scaled post-training rather than a new foundation-model architecture. According to the company, GLM-5.3 delivers roughly a 50% improvement over GLM-5.2 on its internal Z.ai Code Bench.

That makes GLM-5.3 another model well suited to agent-style workloads where the model needs to plan, use tools, modify artifacts, and continue working over an extended sequence of steps.

The original AIBase article highlights programming, complex long-horizon engineering, and defensive cybersecurity as key areas where the model stands out.

Qwen3.8-Max Completes the Three-Model Lineup

QwenWork had already integrated Alibaba’s own Qwen3.8-Max, giving the platform a third flagship option.

Alibaba officially launched Qwen3.8-Max in early August and describes it as the most capable model in the Qwen family to date.

The model uses a sparse Mixture-of-Experts architecture with 2.4 trillion total parameters and 95 billion active parameters, while supporting a context window of up to 1 million tokens.

Alibaba positions Qwen3.8-Max around several areas that overlap directly with office-agent workloads:

• Coding
• Professional work
• Research
• Long-horizon tasks
• Multimodal understanding
• End-to-end completion of complex assignments

This makes the three-model lineup less about offering three interchangeable chatbots and more about giving one agent platform access to several different frontier-model profiles.

Users can choose among Qwen3.8-Max, GLM-5.3, and DeepSeek V4 Pro without leaving the QwenWork environment.

QwenWork Is Moving Toward a Multi-Model Agent Strategy

QwenWork entered public beta on August 3, 2026.

Alibaba says the platform combines the core capabilities of three earlier AI agent products:

• QoderWork
• MuleRun
• Wukong (悟空)

The resulting product is designed to bring together desktop agents, cloud agents, and enterprise collaboration agents in one workplace environment.

Alibaba’s official product materials describe QwenWork as an all-in-one AI office platform capable of producing and editing documents, spreadsheets, presentations, and HTML pages while also handling multi-step workflows, multimodal content, data aggregation, skills, and enterprise collaboration.

The product is available through both a web interface and a desktop client. Alibaba has also said that QwenWork will be integrated more deeply with DingTalk and that a standalone mobile app and international version are planned.

The addition of GLM-5.3 and DeepSeek V4 Pro adds another layer to that product strategy.

Instead of presenting QwenWork only as an Alibaba agent powered by Alibaba models, the platform is increasingly acting as a multi-model office-agent interface.

That approach lets the product layer sit above the foundation-model layer. The user chooses the workspace and workflow first, then selects an appropriate model for the task.

For enterprise AI products, this can be useful because different models may have different strengths in coding, document work, long-context processing, cost, latency, or specialized agent tasks.

More Choice in China’s Enterprise Agent Market

The AIBase article places this update in the context of an increasingly competitive Chinese office-agent market.

Products such as Tencent WorkBuddy and ByteDance TRAE Work are also competing to become the interface where users delegate multi-step knowledge-work tasks to AI agents.

QwenWork’s latest move adds another dimension to that competition.

The product is no longer differentiated only by its own agent features or by access to Alibaba’s Qwen models. By adding current flagship models from Z.AI and DeepSeek, QwenWork is broadening the model choices available inside the same workflow environment.

That could be particularly relevant for enterprise users who do not want to rebuild an entire workflow every time a stronger model appears.

A multi-model agent layer can, in principle, make model upgrades less disruptive: the surrounding workspace, files, skills, and business workflows remain in place while the model behind a task can change.

The practical value will still depend on how QwenWork routes tasks, exposes model differences, handles pricing and quotas, and maintains consistent tool behavior across different foundation models.

For now, the significance of the update is straightforward: QwenWork now brings together three of the most prominent newly released Chinese flagship models inside one office-agent product.

常见问题

What is QwenWork?

QwenWork, also known in Chinese as 千问办公, is Alibaba’s all-in-one AI office agent platform. It combines desktop, cloud, and enterprise collaboration agent capabilities and can produce office documents, work with data, create web content, and execute multi-step tasks.

Which flagship models are available in QwenWork?

According to the August 2026 update, QwenWork supports Qwen3.8-Max, GLM-5.3, and DeepSeek V4 Pro in its frontier-model lineup. The two newest additions are GLM-5.3 and DeepSeek V4 Pro.

What is DeepSeek V4 Pro’s context window?

DeepSeek’s official API documentation lists DeepSeek-V4-Pro-0813 with a 1 million-token context length and a maximum output of 384K tokens. It also supports tool calling, thinking mode, JSON output, and several API interfaces.

What is GLM-5.3 designed for?

Z.AI positions GLM-5.3 as a frontier model for coding and agentic workloads, with particular emphasis on long-horizon engineering and cybersecurity capabilities. Z.AI reports about a 50% improvement over GLM-5.2 on its internal code benchmark.

What is Qwen3.8-Max?

Qwen3.8-Max is Alibaba’s flagship Qwen model released in August 2026. It uses a 2.4-trillion-parameter sparse MoE architecture, supports up to a 1 million-token context window, and is designed for coding, professional work, research, multimodal tasks, and long-horizon execution.

Can individuals use QwenWork?

Yes. Alibaba opened QwenWork’s public beta to both individual and enterprise users in China. The official website provides a web version and desktop-client downloads.

Does QwenWork only use Alibaba models?

No. The addition of GLM-5.3 from Z.AI and DeepSeek V4 Pro from DeepSeek shows that QwenWork is adopting a broader multi-model strategy rather than limiting its frontier-model tier to Qwen models.

Is a QwenWork mobile app available?

Alibaba has said that a standalone mobile app and an international version are planned. Readers should check QwenWork’s official website for the latest availability because product rollout schedules can change.

相关工具

• QwenWork: Alibaba’s AI office agent platform for documents, data, web deliverables, multimodal work, skills, and enterprise workflows.
• DeepSeek API: Official developer documentation for DeepSeek models including DeepSeek V4 Pro.
• Z.AI: The official platform behind the GLM model family, including GLM-5.3.
• Alibaba Cloud Model Studio: Alibaba Cloud’s model platform for accessing Qwen models such as Qwen3.8-Max.

Related Links

• QwenWork Official Website: The official QwenWork product site with web access, downloads, features, and enterprise information.
• Alibaba Announces QwenWork Public Beta: Alibaba’s official August 3 announcement describing the platform and its integration of QoderWork, MuleRun, and Wukong.
• Qwen3.8-Max Official Announcement: Alibaba Cloud’s official release overview for Qwen3.8-Max.
• Qwen3.8-Max Model Documentation: Official Model Studio documentation for the Qwen3.8-Max model.
• GLM-5.3 Official Release: Z.AI’s official introduction to GLM-5.3, including coding and post-training improvements.
• DeepSeek Models and Pricing: Official model specifications, context limits, output limits, features, and API pricing for DeepSeek V4 Pro.
• Qwen3.8-Max: A New Bar for Coding and Cowork: Qwen’s official technical introduction to its August 2026 flagship model.

Summary

QwenWork has added GLM-5.3 and DeepSeek V4 Pro to its frontier-model lineup, joining the previously integrated Qwen3.8-Max.

The update gives Alibaba’s office-agent platform access to flagship models from three major Chinese AI developers while keeping the surrounding workspace and agent experience in one product.

For QwenWork, the shift is as important as the individual models themselves: the platform is moving from a single-model product toward a multi-model agent layer designed around completing workplace tasks.

The latest integration makes model choice part of the QwenWork workflow rather than forcing users to choose a separate AI product for every flagship model.