For months, many leading artificial intelligence developers in the U.S. have been racing to develop sophisticated AI agents that can carry out more complex tasks on a user’s behalf.
Now, a lesser-known Chinese startup is claiming to have vaulted ahead of them.
Manus AI last week launched a preview version of what it called a general AI agent capable of screening resumes, creating trip itineraries and analyzing stocks in response to basic instructions from the user. The company said its service performed better on some fronts than OpenAI’s Deep Research, another recently released agent.
While some AI agents require a certain amount of hand-holding and supervision, Manus co-founder and chief scientist Yichao Ji said its product is “truly autonomous.” A slick video demonstration from the company quickly went viral, prompting a scramble for a limited number of invites to try out the tool.
The early frenzy has earned Manus comparisons to DeepSeek, the Chinese upstart that rattled Silicon Valley in January by releasing a competitive AI model that it claimed to have developed for a fraction of the cost that U.S. rivals have spent on their own technology. Like DeepSeek, Manus is once again sparking questions about the U.S. lead on artificial intelligence—this time in a product category that U.S. tech companies see as a key investment area.
So far, however, the initial reactions from Manus users have been mixed. Derya Unutmaz, a professor at the Jackson Laboratory who researches cancer immunotherapy, praised the tool for “great quality outcomes” even though “it takes longer than OpenAI’s Deep Research to process the tasks.”
Others have complained that the service is too slow and sometimes crashes before completing tasks, likely due to the company’s limited computing resources. Some users have also found it made factual mistakes.
“Manus is actually a half-finished product,” said Yiran Chen, an electrical and computing engineering professor from Duke University who has tried out the service. The startup likely hopes “that by being first, they can attract investors, despite the product not being fully developed yet.”
While speculation on social media has run rampant about the company, some details about its development have emerged. Ji said Monday on X that Manus’s offerings are based on Claude and fine-tuned versions of Alibaba Group Holding Ltd.’s Qwen models, although it is unclear to what extent it simply refines and builds on top of those technologies. Still, Manus’s approach is likely to cost much less than building cutting-edge systems from scratch, which can cost tens of millions of dollars.
The company behind Manus, Butterfly Effect, has raised more than $10 million in financing, several China-based media outlets have reported, including from prominent VC firms ZhenFund and HongShan. Butterfly Effect is currently advertising 20 openings on China’s leading job platform, Zhipin.com, with annual salaries as high as 560,000 yuan ($77,130) for an engineer.
Unlike DeepSeek and some U.S. companies, Manus did not publish detailed papers alongside its release to provide details about how the technology was developed. It also hasn’t publicly released any of the code or weights associated with its AI for people to use independently.
Manus’s co-founder and Chief Executive Officer Xiao Hong is seeking to compete globally and not just in China, he said in an interview with Tencent News in February.
Those trying Manus can select from one of two options: a standard mode or a high-effort mode, the latter of which takes more time to process requests. Yong Qian, founder of OmniEdge Inc., said Manus “creates a task list, performs each step one by one, fixes any problems it encounters and asks questions if it needs new commands.” Then it offers up a response.
In recent months, OpenAI, Anthropic and other AI companies have released agents that can use a person’s computer to browse the internet, conduct online research and complete various multi-step tasks. But some early Manus users thought the tool rivaled what’s currently on the market.
“I’ve tried hundreds of AI tools. I keep trying new tools every single day,” said Ashutosh Shrivastava, a Bangalore-based software developer, who used the service to build a website and a game. “I haven’t seen anything like Manus.”
