Chinese Team Unveils AI Agent, Manus

Many people spent the night of March 6th searching for invitation codes to Manus, Monica’s AI Agent, which was launched at the same time as Apple’s new product.

According to their team Manus, is a truly independent AI agent capable of handling a variety of complex and dynamic tasks. Manus is different from traditional AI assistants in that it not only gives suggestions or answers, but also completes the task.

Manus uses a multisig system, which is driven by several independent models. The official plan is to release some of these models later this year, especially the ‘postering (inferences)’ part of Manus.

A four-minute demo was also used to gain attention. In these cases, Manus completed the entire process autonomously from planning to execution. This demonstrated genuine Agent capabilities, rather than simple assistant features.

As an example, the first step involved screening 15 resumes for candidates who would be suitable for a position as a reinforcement learning algorithm developer. Manus showed qualities similar to a human assistant, manually unzipping and reviewing each resume, page by page, recording key information. The official team said that what was demonstrated is only a small portion of Manus’s capabilities.

Users were amazed when the AI Agent, after a long chain thought and tool usage, produced a complete, professional result.

To ensure reproducibility, Manus was tested using configurations identical with its official version.

According to the official site, Manus achieved a new state-of the-art (SOTA), across all three levels of difficulty in the GAIA benchmark tests (which assesses general AI assistants’ ability to solve real-world issues).

Xiao Hong is the founder of Manus AI. He graduated from Huazhong University of Science and Technology in Huazhong with a degree of software engineering. After graduation, he launched Yi Ban Assistant and Wei Ban Assistant in 2015, serving more than 2 million B-end customers and securing investment from TencentZhenFund and others.

He also developed Monica, a product marketed as an ‘All-in-One’ AI assistant, which was launched initially as a browser plug-in. Integrating large-scale models (such Claude 3.5 or DeepSeek) into mainstream software. Monica offers functions like chat, translation, copywriting, etc. Monica began by focusing on overseas markets. It reached a user base exceeding one million users and became a leader in the AI plugin market. The Chinese version of Monica, which began internal testing in February of this past year, is now available to domestic users for free. This version is based on the DeepSeek R1 & V3 models and has deep reasoning capabilities, memory functions and real time internet search.

Manus adheres a technical philosophy that is a little different from the mainstream, namely ‘less structure, more intelligence’. They believe that if the data is good enough, the model powerful enough, the architecture flexible enough, and engineering robust enough, capabilities like computer use, deep-research, and coding agents, will naturally emerge without needing to design specific product features.

OpenAI, one of the leading technology companies in the AI industry, announced on March 6th that it would charge $20,000 per month (approximately $145,000 RMB) for AI Agents at the doctor level. This is primarily targeted to enterprise users in data-intensive fields such as finance and healthcare.

Sign up for 5 articles free per month !

www.aiobserver.co

More from this stream

Recomended