Leading ChatGPT Agent by Three Months: Why National-Level Products Are Entering the Universal Agent Arena

Deep News
Aug 20

The global universal agent leaderboard has been refreshed once again. On August 18, Baidu Wenku and Baidu Drive jointly launched "GenFlow 2.0," claimed to be the world's first full-platform universal agent. In a post-launch interview, Wang Ying, head of Wenku and Drive business units, stated: "What Wenku and Drive deliver are ready-to-use products."

As she mentioned, GenFlow entered the market four months ago with an immediately usable approach. The latest Wenku GenFlow 2.0 continues this pragmatic style, launching immediately across Baidu Wenku's web and app platforms without requiring invitation codes.

Moreover, GenFlow 2.0 can coordinate 100+ agents running tasks simultaneously, completing over 5 complex tasks in parallel within just 3 minutes, achieving speeds 5-10 times faster than similar products.

**3-Minute Completion of 5+ Complex Tasks: Multi-Agent Collaboration No Longer "Crashes"**

The memory of viral universal agents being "hard to get" and "sold at premium prices" remains fresh in users' minds, but a demystification has already begun. Ideally, users should only need simple natural language interactions to have universal agents coordinate multi-agent systems and complete complex tasks. However, as users began hands-on testing, problems gradually emerged.

Many users reported after practical experience that universal agents suffer from issues requiring repeated input adjustments, low task delivery quality, excessive waiting times, hollow and incorrect answers, and inability to edit content. In various forums, "Why do Multi-Agents always fail?" has become a hot discussion topic.

This is supported by data: a team from UC Berkeley and other institutions surveyed failure rates of five overseas Multi-Agent systems, with one major tool showing a failure rate as high as 86.7%. This indirectly indicates that many existing universal agents cannot accurately capture requirements and execute tasks, remaining far from the ideal state.

These issues reflect that many universal agents are limited by system design flaws, data gaps, insufficient knowledge base coordination, and other factors, preventing them from realizing their full potential and leaving many users with a "first impression" of inadequate AI capabilities.

When communicating with media, the Baidu Wenku and Drive teams shared that this is precisely the challenge GenFlow 2.0 aims to overcome.

To evaluate whether Wenku GenFlow 2.0's performance meets expectations, a comprehensive enterprise research task targeting global GPU chip leader NVIDIA was input, requiring understanding of the company's development history, annual financial status, and business layout, with results output in different file formats.

The execution process showed that Wenku GenFlow 2.0 immediately broke down the task into 8 subtasks, used deep search tools to collect core information, and completed execution within minutes. The generated content was fundamentally accurate and formatted according to enterprise research requirements.

Another task generating teaching materials for middle school physics teachers (containing 5 subtasks) was completed within 3 minutes. Official data shows GenFlow 2.0's operating speed is 10 times that of similar products, achieving minute-level "parallel" work and delivery.

Behind the speed improvement, Wenku GenFlow 2.0 adopts self-developed Multi-Agent infrastructure rather than mainstream agent serial workflows. To enhance interactive experience, the Wenku app has upgraded from conventional "waterfall-style" workflows to "parallel-style" workflows.

Additionally, during the generation process, Wenku GenFlow 2.0 supports user "intervention" at any time during tasks. Users can request pauses, ask follow-up questions, modify thinking content, upload reference files, etc., in the GenFlow 2.0 chat window according to scenario needs. GenFlow 2.0 adjusts generated content in real-time based on the latest requirements.

In terms of delivery capabilities, beyond the PPT, documents, and images experienced, GenFlow 2.0 also covers multi-modal content including research reports, video picture books, posters, charts, HTML, code, games, and websites.

Behind the full-modal generation capability, Wenku GenFlow 2.0 can invoke an "AI expert team" composed of 100+ multi-modal agents, comprehensively meeting user needs. All agents have been validated by hundreds of millions of Wenku and Drive users.

For example, Wenku's PPT Agent has achieved global first place in user visits with an adoption rate approaching 90%; the research report agent has pioneered minute-level generation of tens of thousands of words of professional reports while generating professional-grade visualization charts.

Wang Ying revealed that to ensure user experience, GenFlow 2.0 currently integrates primarily Baidu-developed agents, but third-party agents will become increasingly diverse as the product develops.

**Two Years of Comprehensive AI Reconstruction: Wenku and Drive Demonstrate Renewed "Growth Power"**

To understand the reasons behind GenFlow 2.0's performance improvements, beyond the aforementioned functional design, one cannot ignore its foundation in Baidu's ecosystem resources. Starting in 2023, Wenku and Drive began a two-year comprehensive AI reconstruction.

At GenFlow 2.0's launch event, the core team introduced that GenFlow's positioning is to provide an efficient AI expert team that "remembers its mission, has clear goals, follows commands, and can win battles." The development team made substantial investments in the technical foundation.

For instance, at the model level, Wenku and Drive have consistently adopted MoE (Mixture of Experts) architecture since AI reconstruction began, enabling different models to be called based on different tasks and steps—this is also the mainstream architectural choice for current large models. Starting in 2024, many large model players began transitioning to MoE.

Technically, GenFlow 2.0 has built its own Multi-Agent engine and actively optimized contextual engineering to construct an information ecosystem for models. Through dynamic hybrid inference, multi-mode intelligent scheduling, public-private domain knowledge enhancement, and full-modal rendering and editing, it achieves one-stop end-to-end delivery of complex tasks. It has also built a foundational platform including capabilities such as full-lifecycle message buses, multi-modal data understanding, short-medium-long memory centers, and multi-level risk control systems, striving to achieve optimal solutions for performance, effectiveness, and scalability.

At Baidu Create 2025 Developer Conference held in April this year, Baidu Wenku and Baidu Drive jointly launched the content operating system "Cangzhou OS" and debuted "GenFlow 1.0" based on this system.

Currently, Baidu Wenku possesses over 1.4 billion professional content resources, with AI monthly active users exceeding 97 million and hundreds of AI capabilities including smart PPT, smart documents, and AI picture books. Baidu Drive has cumulatively served over 1 billion users, with monthly active users exceeding 200 million and AI monthly active users surpassing 80 million.

Furthermore, GenFlow 2.0 is compatible with MCP protocol, enabling flexible integration with third-party service ecosystems and supporting manufacturers, enterprise users, intelligent applications, and developers to connect and provide users with richer services.

Currently, Honor has become the first hardware manufacturer globally to access the MCP ecosystem, natively integrating GenFlow 2.0 into Honor's intelligent assistant YOYO. Through this integration, Honor MagicOS users can access personal cloud knowledge bases and Wenku professional documents with one click, achieving experiences including cloud search, content sharing, internet search, image understanding, file summarization and Q&A, and Wenku PPT generation, realizing system-level native scheduling between AI agents and hardware manufacturers.

However, this is still far from reaching the ultimate goal of universal agents. Before that arrives, Baidu Wenku and Drive will continue their efforts. At the event, Wenku and Drive announced upgraded developer benefits, newly opening and optimizing over 100 capabilities while providing free quotas for all capabilities to empower more developers.

As enterprises resonate with market pulse, and as user curiosity fades while questions about value become unavoidable challenges for all market players, Baidu's AI reconstruction continues to deepen as the global large model application market enters deeper waters.

The universal agent market doesn't yet have standard answers, but only through continuous exploration and investment can people's imagination of "future work methods" and "super productivity" become reality. Including Baidu, the market continues its rapid sprint, but many interesting changes are already occurring.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10