Manus, the Viral Chinese AI Agent Some Call 'the Second DeepSeek'

A new player has entered the conversation around Chinese AI, and this time it's not a language model but an agent. Manus, built by Wuhan-based startup Butterfly Effect, has spread like wildfire on social media this week, both inside and outside China. Among those praising its capabilities are Twitter cofounder Jack Dorsey and Victor Mustar, product lead at Hugging Face. Some are already calling it "the second DeepSeek," a nod to the model that stunned the industry just weeks ago with its unexpected capabilities and its Chinese origins.

What Manus Actually Is

The company bills it as the world's first "general" AI agent. Unlike a chatbot such as ChatGPT or DeepSeek itself, which rely on a single family of language models built for conversation, Manus combines several AI systems operating independently to carry out tasks start to finish without constant oversight. Among the models it draws on are Anthropic's Claude 3.5 Sonnet and fine-tuned versions of Alibaba's open-source Qwen.

That multi-agent architecture, with several systems working in parallel, is what sets it apart from autonomous research tools like OpenAI's DeepResearch, which also browse the web and break tasks into steps but take a more single-model-centric approach.

Access Is Nearly Impossible, Yet Hype Keeps Building

Despite all the media buzz, very few people have actually gotten to try it. Fewer than 1% of those on the waitlist have received an invite code, as has been confirmed in recent days. The exact size of that waitlist isn't known, but Manus's Discord channel already has more than 186,000 members — a number that hints at just how much interest it has generated in only a few days since launch.

Butterfly Effect isn't an unknown startup: in 2023 it launched Monica, an AI assistant also aimed at a global audience, with English as the default language. Manus follows that same international ambition, with a minimalist design reminiscent of ChatGPT or DeepSeek, and an interface featuring a window dubbed "Manus's Computer," where users can watch in real time what the agent is doing and step in at any point in the process.

Why It Matters Beyond the Hype

The Manus phenomenon arrives just weeks after DeepSeek rattled the industry with a low-cost, high-performance model that called Silicon Valley's technological edge into question. The comparison between the two products is imperfect — Manus isn't a foundation model but an orchestration layer built on top of existing models — but the underlying takeaway is the same: Chinese AI companies aren't just replicating their Western competitors' base models. They're setting the pace in the practical adoption of autonomous agents, one of the frontiers drawing the most attention in the industry over the past year.

The promise of an agent capable of carrying out complete tasks — from business strategy to apartment hunting to personalized meditation sessions, according to the examples the company itself showcases on its website — taps into a trend that Google, OpenAI, and Anthropic have been exploring for months: the shift from conversation to execution. That this leap is arriving with viral force from a Chinese startup in Wuhan, rather than from one of the big American labs, is the part of the story drawing the most attention this week.

With access still limited to a handful of users, it's too early to say whether Manus delivers on its promises or whether this is just another episode of runaway hype in an industry well accustomed to them. What is clear is that the bar for what's expected of an autonomous agent has risen another notch — and that the center of gravity in this race is, once again, shifting toward China.