Following the release of the R1 model by the Chinese AI startup DeepSeek that shook the AI community, yesterday another Chinese team, Monica.im, unveiled their AI agent "Manus", which has become a sensation, claiming to be the world's first general-purpose AI Agents.
Unlike OpenAI's Deep Research, which is focused solely on analysis and research output, the key feature of Manus is its ability to complete more complex tasks, such as:
- Resume Screening: Screening suitable candidates from 15 resumes and generating an Excel sheet with rankings.
- Comparative Property Filtering: Researching (through web browsing and API usage) safe neighborhoods and quality schools for New York home buyers, and using Python to calculate budgets and filter properties.
- Dynamic Website Creation: Designing content structures and implementing interactive websites on specific topics, whether it's games or educational websites.
- Online Learning Resource Development: Directly designing online websites on specific topics to provide relevant teaching.
To see more Manus use cases, visit the official website and join the official Discord...
Manus is Deep Research + Operator + Computer Use
When Manus runs, it first analyzes the requirements and clearly records the entire thought process in a separate document, then executes step-by-step. In this process, it may call various APIs to collaborate on the task, or it may launch a browser to simulate human-web interaction and exploration, ultimately organizing, analyzing, and outputting the user's desired format, whether it's a document, spreadsheet, App, website, or PPT.
HashMatrix co-founder Pang Shunxin described Manus as a combination of Deep Research + Operator (OpenAI's daily task assistant) + Claude Computer Use (directly allowing AI to control your computer). In simple terms, it's more user-friendly, allowing users to achieve their goals more quickly.
The release of Manus has shown the market that it can indeed replace many daily life and work tasks for people, fully embodying its website slogan: "Hand it to Manus." Furthermore, the term "Manus" is derived from Latin, meaning "hand," symbolizing Manus' ability to complete various complex tasks.
GAIA Surpasses OpenAI
According to the website, Manus has demonstrated the most advanced general-purpose AI assistant capabilities in the GAIA Benchmark (a general-purpose AI assistant evaluation benchmark), surpassing OpenAI's Deep Research and previous best performances across all difficulty levels, with significant progress in solving high-difficulty problems (Level 3).
Invitation Code Soars to TWD 220,000
However, using Manus requires an invitation code, and its popularity has caused the registration page on the official website to crash, sparking a frenzy in the community to obtain the invitation code, with prices being driven up to between RMB 999 and RMB 50,000 (approximately TWD 4,400 to TWD 220,000).
Currently, the official waitlist link has been opened. Manus AI partner Zhang Tao explained that the invitation code mechanism is due to limited server capacity, as the original plan was to prepare according to the industry standard for releasing a Demo (concept demonstration), but the market demand turned out to be much larger than expected.
Is it Really a "General-Purpose" AI Agents?
However, some experts still have doubts about whether Manus is truly a "general-purpose" AI agent. MyShell co-founder Zengyi Qin pointed out in a post that Manus is an excellent product, but not a technological breakthrough, mainly because:
Constrained Execution: Manus relies on pre-defined workflows to complete tasks, which can improve short-term system efficiency, but in the long run, it limits its ultimate performance.
Restricted Operating Environment: Manus can only run at the browser and OS level, unable to use software beyond its tool range. For example, it cannot operate PowerPoint to automatically create presentations.
True Technological Breakthrough Standard: A true breakthrough should be a fully open OS-level AI assistant that can perform most tasks like a human and operate various software, not just run within a specific environment.
This commentary suggests that while Manus is powerful, it is still constrained by a specific environment and has not yet truly achieved the vision of a "general-purpose AI agent".
