- The release of the Manus general AI agent from China is being touted as the "second DeepSeek moment."
- It claims to achieve a wide range of tasks and can browse the web, generate a detailed plan, operate a computer, and run code to accomplish tasks.
- It has been revealed that Manus AI is powered by Claude 3.5 Sonnet, Browser Use, and several fine-tuned versions of Qwen models.
The release of China’s Manus AI agent is being hailed as the “second DeepSeek moment.” Unlike standalone AI agents, Manus is a general AI agent that can accomplish a broad range of tasks you throw at it. However, if you’re curious how Manu AI works and who the people behind it are, then keep reading. To better understand China’s latest AI agent, go through this detailed explainer on Manus AI, its technologies, and more.
What is China’s Manus AI Agent?
Manus is a general AI agent that can autonomously browse the web, run code, and interact with a computer to complete a wide range of tasks. It has been developed by a Chinese company called Butterfly Effect, which is headquartered in Wuhan.
Butterfly Effect is a relatively small company with only dozens of employees, based out of Wuhan and Beijing. Currently, Manus AI is in closed beta phase, and it’s only available via invitation.
What is unique about Manus is that it has integrated a lot of tools into a single workflow. For example, OpenAI’s Operator agent is a Computer-using agent that can interact with a cloud computer, however, it’s available as a separate product. Similarly, OpenAI’s Deep Research agent can browse the web to generate a comprehensive report. Again, it’s a separate product.
Now, what Manus AI is doing is clubbing all these agents and tools to create a single “general” AI agent that can perform deep web research, find information, generate a detailed plan, operate a computer, and run code in an isolated session.
OpenAI has also said that GPT-5 will be a unified system, integrating ChatGPT’s tools into a single AI system. However, China’s Manus AI has taken the lead in this direction before ChatGPT.
What Powers the Manus General AI agent?
Now, this begs the question, what powers the Manus general AI agent? The chief scientist of Manus, Yichao ‘Peak’ Ji revealed in an X post that Manus uses Anthropic’s Claude 3.5 Sonnet model and various fine-tuned versions of Alibaba’s Qwen models.
In addition, it uses the open-source Browser Use AI agent (GitHub) to interact with websites in a web browser. There are also reports saying that Manus has access to 29 different tools.
Manus is internally testing Anthropic’s latest Claude 3.7 Sonnet unified model, which will improve the agent even further. In addition, Reuters reports that Manus has partnered with Alibaba’s Qwen team to expand its general AI agent.
What Can Manus AI Agent Do?
While we don’t have access to Manus AI agent right now, the company has listed various use cases on its website. It can book flight tickets, reserve a table at a restaurant, analyze stocks and earnings reports of companies, perform data analysis, and more. Moreover, Manus can sort through resumes and analyze all the files, generating a report with candidate profiles and ranking suggestions.
You can also download files in Excel or Word documents on your local computer. By the way, the Manus AI agent has access to its own computer environment where it can preview files, interact with them, use a web browser, run Python code, and more. It breaks down the tasks into small steps and then goes to the web to accomplish them one by one.
You can see what the agent is doing and also intervene in case the agent takes a wrong turn. Once the task is done, you are notified on your computer or smartphone.
Manus AI Benchmark Performance
The Manus AI agent has been benchmarked on GAIA (General AI Assistants), which is a rigorous benchmark that challenges AI agents to solve real-world questions. It requires “a set of fundamental abilities such as reasoning, multi-modality handling, web browsing, and generally tool-use proficiency.”
On GAIA, Manus achieved the highest score of 86.5% in Level 1 tasks, outperforming OpenAI’s Deep Research agent, which scores 74.3%. By the way, we humans score around 92% on GAIA tasks. Manus did better than OpenAI’s Deep Research agent on Level 2 and Level 3 tasks as well.
Overall, the Manus general AI agent marks a promising step toward an agentic future. While some of the hype may be blown out of proportion, the Manus team has done a good job integrating a lot of tools. Remember that Manus is still in closed beta and the team has assured that it will be improved significantly before the AI agent becomes available to the general public. It looks like OpenAI has a new competitor in the AI agent space, besides DeepSeek, and both hail from China.
