Adapted by:
Javier Pastor
Senior WriterComputer scientist turned tech journalist. I've written about almost everything related to technology, but I specialize in hardware, operating systems and cryptocurrencies. I like writing about tech so much that I do it both for Xataka and Incognitosis, my personal blog.
Karen Alfaro
WriterCommunications professional with a decade of experience as a copywriter, proofreader, and editor. As a travel and science journalist, I've collaborated with several print and digital outlets around the world. I'm passionate about culture, music, food, history, and innovative technologies.
A purportedly new DeepSeek-style AI phenomenon is making waves. Manus, a theoretical AI agent, has sparked excitement with its striking capabilities. However, skepticism is growing online about its actual potential.
Not an AI agent. Manus’s website describes it as “a general AI agent that bridges minds and actions: it doesn’t just think, it delivers results.” In reality, it functions more like a deep research tool—similar to OpenAI’s deep research mode—analyzing user-provided data. Manus claims to create detailed travel itineraries, assess stock values, and generate class presentations on topics like momentum conservation.
Click on the image to view the original post on X.
High expectations. A Hugging Face product manager called Manus “the most impressive AI tool I’ve ever tried.” Its official Discord server has attracted nearly 170,000 users in just a week. The buzz is undeniable, but as TechCrunch notes, doubts persist.
Built on existing models. Manus isn’t entirely new. Users have pointed out that it appears to rely on existing AI models. Apparently, it’s totally based on Claude 3.7 and Alibaba’s Qwen. Co-founder Yichao “Peak” Yi confirmed that Manus builds on these models rather than being developed from scratch. In a presentation video, Yichao claimed Manus outperformed OpenAI’s Deep Research and Operator in benchmarks. But is that true?
Errors and inconsistencies. AI experts, including Pleias co-founder Alexander Doria, have reported errors and infinite loops when using Manus. Others say it struggles with factual questions and fails to consistently cite sources.
Invite-only beta. Currently, access to Manus is invitation-only. TechCrunch, after gaining access, found the results underwhelming. The reporter’s tests—including booking a restaurant table and purchasing airline tickets—failed.
The Register also tested Manus and was “not at all impressed.” When attempting to generate a Mario clone, the results were poor. Despite its flaws, some users see promise in the product.
Not another DeepSeek—yet. Unlike DeepSeek, which launched as open source software, Manus remains closed. The company may be leveraging invite-only hype while managing infrastructure costs. For now, Manus is generating buzz—but it has yet to prove itself as the next big thing in AI.
Image | Manus
Related | How to Use DeepSeek: 36 Features and Tricks to Get the Most Out of This AI Model
View 0 comments