Artificial Intelligence (AI) Startup Jens Park unveiled ‘Super Agent’ and predicted a full -fledged competition in the final AI agent market. Specifically, he insisted that China’s representative AI agent ‘Manus’ showed performance.
Jens Park announced on April 4 that it has launched a general -purpose AI agent ‘Super Agent’, which might autonomously perform users’ every day work.
Super Agent is a high -performance autonomous AI system designed to quickly handle various actual tasks, which might perform extensive tasks comparable to schedule management, voice call reservations, and customised video production.
Nine large language models (LLM), greater than 80 AI tools, and greater than 10 databases are built. This can be a big difference in size from two models, ‘Claude’ and ‘Q1’ and Manus with some tools.
Within the released demo video, a brilliant agent was introduced to complete the restaurant reservation with a voice call by utilizing the map tool to calculate the walking distance and public transportation channels, and to contemplate allergies and seat conditions. As well as, when a selected cooking video was requested, the function of routinely processing from recipe surveys to video production and insertion of sound effects was also demonstrated.
Experts cited the strength of the super agent as ‘showing the reasoning process to the user.’ The interface that transparently shows which tools you selected is to extend user understanding and reliability.
As well as, the browser -based interface could be used immediately without installation, and the barrier to entry is low since it doesn’t require personal information input. Then again, the competitive product Manus is less accessible since it must link SNS accounts and register a listing of waiting lists.
Technically, it’s also noteworthy that it effectively solved the ‘tool orchestration’ problem that links various external tools and APIs.
To this end, the routing technology that connects the optimal model for every task and the Model Context Protocol (MCP) standard that directly connects external data sources.
As well as, the corporate introduced the recent framework ‘Cotools’ developed by Suzhou University in China, improving the best way LLM uses external tools. Unlike the present prompt engineering or micro -adjustment methods, the tool chain is a technique of putting the essential model in a ‘frozen’ state and training a small module individually for using the tool.
It recorded 87.8%of the AI ​​Agent Performance Evaluation Index, ‘GAIA’ benchmark, and beat Manus AI’s Manus (86.5%) and Open AI’s ‘Deep Research (74.3%)’.
Zen’s Park emphasized that it’s different from competitors comparable to Manus and Open AI’s ‘operator’ by utilizing a big selection of models and direct APIs.
Manus relies on a small variety of models, has less integrated tools, and the operator is proscribed to browser -based work. Nonetheless, the Super Agent said that it provides fast data search and wealthy work execution function by the API -centered approach.
Specifically, he emphasizes that he can immediately utilize super agents in various job groups comparable to teachers, designers, and recruiters, and suggests the potential of competing with the present robotics process automation (RPA) platform or software service (SaaS) solution.
Super Agent On the official website It is accessible and provides a free plan and a paid plan ‘Plus’. The paid plan is accessible for $ 24.99 per thirty days.
By Park Chan, reporter cpark@aitimes.com
