Manus: China’s Latest AI Sensation

Did a wrapper just best OpenAI and Anthropic?

Just as the buzz around DeepSeek was beginning to fade, Chinese AI has made waves again with the AI agent “Manus,” launched on March 6th, 2025. Today, we’re here to unpack the Manus launch, explore the business model of Manus’ parent company, and offer a glimpse into the mind of Xiao Hong 肖弘, the founder behind China’s latest viral AI product.

Source.

What is Manus?

Manus claims to be the world’s first general-purpose AI agent. It ostensibly outperforms OpenAI’s ChatGPT Deep Research on the General AI Assistants (GAIA) benchmark. Currently in beta testing, access is restricted to those with invitation codes, which are reportedly being listed second-hand for 50,000-100,000 RMB (whether anyone actually paid that much is another question). Users report impressive performance in basic tasks, for instance rebooking airline flights, beyond what Anthropic’s Computer Use and OpenAI’s Operator have thus far provided to users. The product is also experiencing slowing response times, hinting that Monica.ai may be struggling to scale up compute to meet skyrocketing demand.

Manus started in 2022 as an AI-powered browser plugin, backed by ZhenFund (真格基金). In 2023, the company secured Series A funding led by Tencent (腾讯) and Sequoia Capital China (红杉资本中国). What began as a simple “ChatGPT for Google” browser plugin has since evolved into a full-fledged AI agent.

Monica, the company that developed Manus, operates from Wuhan, rather than from China's major tech hubs like Beijing or Shanghai. In early 2024, ByteDance attempted to acquire Monica for $30 million, but founder Xiao Hong (肖弘) turned down the offer. ByteDance’s plan was to absorb Monica’s team and technology into its Doubao AI ecosystem, a move that would have diluted Monica’s distinct market position. Instead, Monica closed a new funding round at the end of 2024, reaching an estimated valuation of nearly $100 million.

The exact AI models powering Manus remain unclear. The company claims to use multiple models for different tasks. Notably, when prompted to reveal its own system files, Manus reveals it may be powered by Anthropic’s Claude models — which would make operating in China illegal. This probably explains why Monica’s website appears to be blocked in China.

Edit: confirmed by co-founder.

Anyway, the fact that Manus appears to disclose more than it should hints at broader potential security vulnerabilities.

Who is behind Manus?

Founder & CEO, Xiao Hong (肖弘), is a serial entrepreneur and a graduate of Wuhan’s Huazhong University of Science and Technology (华中科技大学). He first made his mark by building WeChat-related tools as a student, admitting that while his “academic performance was quite poor,” he partnered with more technical classmates to build tools. In 2015, he launched Nightingale Technology (夜莺科技) and created Yiban Assistant (壹伴助手), a WeChat management tool that secured early backing from ZhenFund (真格基金).

By 2019, Xiao saw a bigger opportunity in enterprise WeChat tools and developed Weiban Assistant (微伴助手). His timing was perfect—when rival WeTool (微商工具WeTool) was shut down in 2020, Weiban became the go-to alternative, attracting investment offers from Sequoia Capital China (红杉资本中国) and Youzan (有赞). Eventually, Minglue Technology (明略科技) acquired Weiban, marking Xiao’s first major financial success.

Sensing the potential of large AI models, Xiao left Minglue in 2022 to create Monica.ai, originally designed as a “ChatGPT for Google” browser plugin.

Co-founder & Chief Scientist, Ji Yichao (季逸超) dropped out of high school at 17 to develop Mammoth Browser (猛犸浏览器). His talent caught the eye of Sequoia Capital China’s Zhou Kui (周逵), who introduced him to investor Xu Xiaoping (徐小平). Xu invested 1.5 million yuan, giving Ji complete creative freedom. Recognizing the large potential of LLMs, Ji joined Xiao Hong to start Monica in late 2022.

Interview Quotes

Unlike DeepSeek’s media-shy Liang Wenfeng, Xiao Hong has done a ton of press. Below are selected translations from several in-depth interviews with Monica’s founder and CEO, Xiao Hong, offering insights into his vision, strategy, and the future of AI agents.

The vibe of Xiao Hong’s interviews is distinct from the AGI-driven idealism blended with national pride we’ve seen from the founders of DeepSeek and Unitree. Xiao is pragmatic and focused on profitability rather than research. A newly published three-hour podcast with Xiao opens with offering this piece of advice:

“I remember there was a Northeastern Chinese restaurant near my university. I made enough money to treat my tech club friends to dinner there every day. Here’s a tip for the audience: if you’re in college, take your most talented classmates out for meals as often as you can. If you wait until after graduation to recruit them for your startup, you’ll have to treat them to Michelin-starred restaurants instead.”

In another interview from January 2024, Xiao openly admits that he didn’t initially believe in AI’s potential, and “remained cautious” despite the hype surrounding GPT-3.5 in the fall of 2022. He describes coming to two conclusions about AI investment, which eventually led him to focus on AI products as opposed to chasing AGI with foundational model research:

"First, I wouldn’t consider working on big models without sufficient business scale. Second, I believe that in China, big model services will eventually integrate fully with cloud computing. I’ve discussed this with our CTO and believe that cloud computing companies will provide customized deployment services, so we don’t need to dive into that ourselves."

…

"I focused more on what big models could do, and what kind of applications I could build with them. In the beginning, many people were financing based on concepts, but by the second half of the year, both domestic and international, there was much less of that. Everyone was returning to business rationality, focusing on finding PMF (Product-Market Fit). By February of 2023, I had a conversation with an investor focused on big models, and no matter how I asked, they refused to talk about products. They weren’t discussing technology or plans. By March, the product’s valuation plummeted. People realized that simply building a single application based on big models might not work, and that’s when the consensus started forming: either focus on technological breakthroughs or work on relatively closed-loop application scenarios."

…

"In March and April of 2023, the fastest-growing product outside of ChatGPT globally was Poe. It was essentially a shell around a big model, and I told investors that if you can perfect the shell, that’s still a big deal. So we decided to do it too, and instead of resisting the demand, we decided to embrace it. In the first half of 2023, Monica integrated all the major models because that’s what the users wanted, and we started by doing that, figuring out how to find more use cases step by step."

Monica’s business model focuses on catering to the overseas market, which likely explains why their website is devoid of any reference to being based in China. Besides English, Monica’s website has dedicated versions in traditional and simplified Chinese, as well as Russian, Ukrainian, Bahasa Indonesian, Persian, Arabic, Thai, Vietnamese, Hindi, Japanese, Korean, and a slew of European languages.

In Xiao’s words, “We chose to target the overseas ToC market because I felt it was a larger, more commercially viable market. The domestic market seemed a bit more challenging.” Their focus shows: in contrast to DeepSeek’s very low key model launches, Manus’ launch came with a whole sophisticated press push like one you would see out of a YC startup, complete with a very well-produced English-language launch video and early access for select YouTubers and twitter influencers.

International expansion comes with its own difficulties, but Xiao believes those challenges made Monica stronger as a company. He’s recently argued that China would benefit from having more firms look abroad:

Xiao Hong: I think we are still in a great era with many opportunities…. First, it's the AI era. Second, I think we are also in a great era of globalization. I'm not a geopolitical expert, but it seems like every country has its own problems — internally, everyone has their own issues. So overall, the world is becoming more conservative and more isolationist, right? But at the same time, no one wants others to be isolationist; they only want to be isolationist themselves. So, everyone hopes that their own entrepreneurs will think more globally.

I believe China’s entrepreneurs of today should be more aggressive in globalizing. If we see overseas markets as better opportunities, it’s not just about market-driven decisions — we should step into international markets to gain experience. We need to participate in global competition, rather than just competing in the markets we are familiar with.

By the way, this process requires a lot of things. When I started this company, none of our founders had lived abroad for an extended period. Everyone’s English proficiency peaked in high school and declined in college! [

I once joked that if, at the same time, there was another founder who had lived in the U.S. and was placed next to me, I would have chosen to work with that founder myself. But this shouldn’t be the way we compare things — it should be about doing our own thing. Secondly, I had a simple belief at the time: the global market is much bigger, and the market itself will provide the tuition fees for founders to learn. (Laughter)

Besides the AI era, another crucial topic is that we are now thinking about things with a globalized mindset.

Unsurprisingly, this business model also relies on collecting vast amounts of user data. Monica’s free Chrome extension requests expansive access to browser data, including permission to log keystrokes, and Manus “crawls” devices to make suggestions. Xiao is betting that widespread adoption of these products will unlock a treasure trove of monetizable insights.

“The data we collect through our browser plugin is critical. Even though this might not guarantee success, it’s a step in the right direction. The private data we gather, along with contextual information, will help differentiate us from the competition. This is one of the key assets we need to grow.”

Xiao is explicitly describing an intent to build an incumbent advantage on a foundation of user data, and TikTok demonstrates how effective that strategy can be. Reliance on eventual mass adoption could partially explain the high-publicity invite-only launch strategy for Manus (although limited access to compute is also certainly a factor).

That said, he is aware that the politics exist and could get in the way of a Chinese-owned AI agent gaining widespread adoption abroad. He spoke about it in a recent podcast alluding to NeZha 2.

I've come to understand that many things are beyond your control. You should focus on doing well with the things you can control. There are truly too many things beyond our control, like geopolitics. You simply can't control it—you can only treat it as an input, but you can't control it.

I recently asked DeepSeek to explain three terms 贪 (greed), 嗔 (hatred), and 痴 (ignorance) [the ‘three poisons’ of Buddishm recently spotighted in the truly excellent animated movie NeZha 2]. It explained it very well: greed is attachment to favorable circumstances; anger is dissatisfaction with adverse circumstances; and ignorance is not understanding the truth of the world. The "truth of the world" is very profound, so I won't discuss that. But greed and anger are problems many people encounter, as are attatchment to favorable circumstances and dissatisfaction with unfavorable ones.

This business-minded pragmatism shines through in Xiao’s vision for the future — instead of techno-optimist visions of AI-powered drug discoveries or a moon colony staffed by robots, he imagines a world where humanity can return to a glorious past:

“I think that the white-collar lifestyle may be a detour for mankind. If you look at it in terms of a curve or over a longer period, say thousands of years, or even the ten-thousand-year span of human history — it's actually quite rare for people to sit in one place and engage in intense mental work without much physical activity. This is probably only a phenomenon of the past hundred years.

For a longer time in history, maintaining physical health and developing spiritual civilization have gone hand in hand. In ancient times, people also needed spiritual and cultural development, but that involved physical labor as well, which helped strengthen their bodies.

In the past hundred years, however, issues like diabetes and high blood pressure have become widespread because people work in this sedentary way. If we look at humanity as a whole, sitting and working for eight or more hours a day is an anomaly.

If AI can take over these tasks, then people can work fewer hours and go back to living more like they did in the past — focusing more on spiritual and cultural enrichment while also taking better care of their physical health.”

To close, here’s a quote from Xiao about how it feels to live through history:

Xiao Hong: From the time I was born in the 90s until now… there have been significant shifts, from PCs to mobile, then the semiconductor industry, which has been booming behind the scenes, the rise of the internet, and now artificial intelligence. I feel like these opportunities are emerging very intensively. When I watched The Godfather, I realized that if I had lived in that era — it was also a time of change — but if you lived in certain periods, you might not have witnessed such rapid technological progress. Sometimes, when we read history books or ancient texts, it feels like things barely changed, which I think would be a little frustrating!