DeepSeek went viral a few weeks ago, shocking the US tech world and tanking the stock market in the process. The Chinese AI firm surprised everyone with an AI reasoning chatbot as good as ChatGPT o1, but which was trained for a fraction of the cost.
Since then, we’ve learned that hardware will still play a key role in developing advanced (frontier) AI like ChatGPT, but that software optimizations can also help. Also, we learned DeepSeek might have copied ChatGPT outputs as a shortcut to speed up training.
DeepSeek going viral also made the world aware that Chinese AI firms shouldn’t be ruled out despite their inability to purchase the latest chips from Nvidia and other US chip makers. Since the DeepSeek release, we’ve already seen a few impressive text-to-video AI models out of China that aim to compete with Sora. Some might even outperform OpenAI’s model.
The latest viral AI out of China is called Manus, from a company called The Butterfly Effect. Manus isn’t your regular ChatGPT or DeepSeek rival. It’s supposed to be an AI agent that can code on your behalf or browse the web for you. We already have such agents from Anthropic and OpenAI.
OpenAI released two AI agents already, Operator and Deep Research. But only the latter is widely available to premium ChatGPT users. You still need to be a ChatGPT Pro user to access Operator, while Deep Research is available on the Plus plan.
Back to Magnus; I saw news of it making the rounds on social media over the weekend. Apparently, the AI does well in tests, and people are in a hurry to use it. Invites are running low, and they might be selling for thousands of dollars online. That’s according to News, which tested the AI.
However, the Magnus hype seems to be unwarranted. Magnus is still in beta and failed miserably at most of the tasks it was given.
Magnus isn’t completely new. That is, it might not have been trained from scratch. Reports say The Butterfly Effect used existing AI models like Anthropic’s Claude and Alibaba’s Qwen. From there, the Chinese AI firms trained the AI to create research reports, perform actions online, and even code apps and games.
A lead researcher from Manus implied on X that the AI model is superior to Deep Research and Operator, the AI agents OpenAI has released to date. Manus supposedly outperforms rivals in deep research tests on GAIA, a popular benchmark for AI assistants. The test examines the AI’s ability to browse the web and use software.
“[Manus] isn’t just another chatbot or workflow,” Manus engineer Yichao “Peak” Ji said in a video on X. “It’s a completely autonomous agent that bridges the gap between conception and execution […] We see it as the next paradigm of human-machine collaboration.”
People who obtained access to Manus aren’t as amazed in real life. The AI might have gone viral, but this isn’t the next DeepSeek in terms of immediate abilities. It might get there as The Butterfly Effect improves it, but it’s unable to really outperform rivals.
News gave Manus various tests worthy of an AI agent, but the AI mostly failed them. Manus could not order food from a top-rated food nearby. The AI failed to book a flight from NYC to Japan despite working with precise instructions about the type of flight the human wanted.
Manus also failed to make a restaurant reservation. News then asked the AI to build a Naruto-inspired game but got an error after half an hour.
Building a game from scratch is the kind of task I’d expect the AI to fail. But booking a table at a restaurant or ordering food should be very simple. OpenAI’s Operator demo went a lot smoother than that.
The Chinese AI firm provided the blog the following statement on the state of Manus, which offers the excuse you’d expect; Manus is in beta. But aren’t they all? Here’s the comment:
As a small team, our focus is to keep improving Manus and make AI agents that actually help users solve problems […] The primary goal of the current closed beta is to stress-test various parts of the system and identify issues. We deeply appreciate the valuable insights shared by everyone.
All that is to say, Manus isn’t the next DeepSeek from China to shock the US AI landscape. It could become the AI model The Butterfly Effect engineers tease it to be in the near future, but we’re clearly not there. No point in spending thousands of dollars to obtain access to it.
Also, as with DeepSeek, you should be aware of what you’re getting into when using AI made in China, which is subject to local laws and practices, including privacy.
