May 21, 2025, China's leading AI enterprise Kunlun WeiQi officially released Tiangong Super Intelligence, an AI Office intelligent agent based on its proprietary Deep Research technology. Its powerful multi-modal content generation capability and cost, which is only 40% of OpenAI's, have sparked heated discussions in the global AI community. Tiangong Super Intelligence consists of five expert agents and one general intelligence agent, supporting one-stop document generation, PPT creation, table analysis, web design, podcast production, and audio and video content creation. Its performance on the GAIA benchmark achieved a score of 82.42, surpassing OpenAI Deep Research and Manus, ranking first. AIbase comprehensively analyzes the latest social media trends to deeply interpret the technological breakthroughs of Tiangong and their profound impact on the AI productivity tools market.
Tiangong Super Intelligence: Multi-Agent Collaborative AI Office
Tiangong Super Intelligence adopts a multi-agent vertical system architecture, including five expert agents and one general agent, combined with the self-developed Deep Research technology, creating a comprehensive AI Office solution. AIbase learned that each agent has clear functional divisions:
Five expert agents: responsible for document generation, PPT creation, table analysis, podcast production, and web design respectively. Each function integrates Deep Research capabilities, capable of generating professional-level content. For example, the document agent can generate research reports containing charts and citations based on user input; the PPT agent automatically designs content with cool visual elements such as bar charts, line graphs, pie charts, and radar charts, with visual effects comparable to professional design.
One general agent: by accessing dozens of MCP (Model Context Protocol), it supports the generation of various multimedia contents such as images, posters, music, MVs, promotional videos, audiobooks, and picture books, meeting diverse creative needs.
AIbase tests show that Tiangong takes only 5 minutes to generate a business report containing 20 PPT slides and 10 data charts, with clear logical structure and intuitive chart presentation, surpassing traditional tools in both efficiency and quality.
Technical Highlights: Deep Research and Open Source Framework
The core of Tiangong Super Intelligence lies in its self-developed Deep Research model, with a cost only 40% of OpenAI Deep Research but achieving a global highest score of 82.42 on the GAIA benchmark, surpassing OpenAI Deep Research (score around 80) and Manus (score around 78). AIbase analyzed that its technical advantages include:
Efficient Deep Research: Through multi-modal data analysis (text, images, webpages), Tiangong can retrieve and integrate thousands of pieces of information within 10 seconds, generating professional reports with data visualization, with an accuracy rate of up to 90%.
Multi-modal Generation: Supports seamless generation of documents, PPTs, tables, webpages, podcasts, and audio/video content, with generated MVs and promotional videos supporting 4K resolution,媲美 professional production in visual effects.
Open-source Framework: Kunlun WeiQi has encapsulated the Deep Research Agent framework and document, PPT, and table generation capabilities as MCP and open-sourced them on GitHub (kunlun-tech/tiangong) under the MIT license, allowing developers to freely modify and integrate them.
Low-cost Deployment: Thanks to efficient model optimization, Tiangong's inference costs are 60% lower than OpenAI's, supporting complex tasks on a single A100 GPU, suitable for small and medium-sized enterprises and individual developers.
AIbase believes that Tiangong's open-source strategy and cost advantage will accelerate its popularity in the global AI ecosystem, especially appealing to resource-limited developers.
Application Scenarios: Full Coverage from Office to Creativity
Tiangong Super Intelligence's multi-modal capabilities have demonstrated broad application prospects in multiple fields:
Enterprise office: Automatically generates business reports containing data charts, marketing PPTs, and financial tables, reducing **80%** of manual editing time, suitable for corporate strategic planning and client proposals.
Content creation: Podcast agents can generate professional-grade audio programs based on text, while general agents support making MVs and promotional videos, helping self-media and marketing teams quickly produce high-quality content.
Education and research: Generates structured research reports and teaching PPTs, integrating Deep Research functions to automatically search academic literature and data, enhancing research efficiency.
Multi-media production: Generates posters, audiobooks, and picture books through MCP, meeting diversified needs in education, advertising, and entertainment industries.
AIbase predicts that Tiangong's multi-modal generation and cool visualization capabilities will drive its widespread application in corporate digital transformation and creative industries, especially in short video and social media content creation fields.
Community Response: Open-source Ecology and Global Discussions
Tiangong Super Intelligence's release sparked enthusiastic reactions on social media and the Hugging Face community. AIbase observed that its GitHub repository received over 5,000 stars on the first day of release, with developers calling it "the future of AI Office." User feedback emphasized the "cool effects" and professionalism of generated reports and PPTs, particularly outstanding in handling complex industry analyses. Some testers shared generated 4K promotional videos, describing them as "on par with professional studios."
However, the community also pointed out that the initial version of Tiangong may require clearer user instructions when processing unstructured data such as handwritten notes. Kunlun WeiQi responded that it would optimize the model in the coming weeks and plans to further collect community feedback through Kunlun Hackathon (scheduled for June). AIbase believes that Tiangong's open-source MCP framework will inspire global developers to contribute code, accelerating the expansion of the AI Office ecosystem.
Industry Impact: China's Global Competitiveness in AI
Tiangong Super Intelligence's release marks a new breakthrough for Chinese AI companies in global competition. AIbase analyzes that compared to OpenAI's Deep Research and Anthropic's Claude4, Tiangong has more advantages in cost efficiency and multi-modal generation capabilities, especially in enterprise-level office scenarios. Its GAIA score of 82.42 leads globally, further consolidating Kunlun WeiQi's position in the open-source AI field, forming a合力 with the rise of other Chinese AI enterprises like DeepSeek (R1 model) and Alibaba (Qwen2.5-Omni-7B).
However, AIbase noticed that Tiangong's Deep Research function may be limited by network stability when processing real-time web data, suggesting that developers combine local caching to optimize performance. Additionally, the widespread use of open-source MCP requires attention to data privacy and compliance.
Address: https://mcp.so/server/skywork-super-agents/Skywork-ai