In just a week, over 2 million people have signed up for the waiting list to get access to Manus AI, a general AI agent from China. Many are dubbing Manus AI the “second DeepSeek moment,” as it is currently in a closed beta phase that requires an invitation for access.
The excitement surrounding Manus AI is palpable, with many labeling it a “game-changer” and likening it to OpenAI’s Deep Research agent. This buzz is especially significant as China continues to roll out new artificial intelligence innovations at lower costs. However, it’s important to consider that the excitement might be overstated, partly fueled by AI influencers making bold claims across social media. Although Manus AI shows promise, I believe it’s not quite a groundbreaking development.
Why Manus AI Falls Short of a Breakthrough
DeepSeek is considered a breakthrough because it successfully mimicked OpenAI’s reinforcement learning (RL) method, achieving performance comparable to advanced reasoning models. Remarkably, the DeepSeek team accomplished this on a limited budget compared to OpenAI’s training expenditures. They also went on to open-source their GRPO training method, which enabled other labs to train top-tier reasoning models.
These achievements represented true innovation, especially given the GPU restrictions imposed by the United States. In stark contrast, Manus AI combines Anthropic’s Claude 3.5 Sonnet model with several fine-tuned Qwen models and is dependent on the open-source Browser Use project.
While Manus AI offers better integration and tooling, the real achievement lies in developing cutting-edge models optimized for agentic tasks. Anthropic’s Claude 3.5 Sonnet is recognized as one of the top models for handling such tasks, including coding. Interestingly, the Manus team is currently testing the new Claude 3.7 Sonnet unified model and has reported encouraging results.
In essence, the core challenge will remain about building capable AI models well into the future. However, it’s worth acknowledging the Manus AI team for effectively linking multiple tools and platforms to achieve their objectives. This represents a promising initial effort towards a more agentic future.
Challenges Faced by Manus AI Agent
Although we don’t yet have access to Manus AI, some early users on X have shared their experiences. Biomedical scientist Derya Unutmaz posted comparisons between Manus and OpenAI’s Deep Research agent. He reported that while Deep Research completed its task in just 15 minutes, Manus took 50 minutes and still failed to finish. He also noted that Manus lacks the source-referencing capability that Deep Research has.
Deep Research finished in under 15 minutes. Unfortunately, Manus AI failed after 50 minutes at step 18/20! 😑 It was performing quite well—I was watching Manus’ output & it seemed excellent. However, running the same prompt a second time is a bit frustrating as it takes too long! https://t.co/bGtmOI65CP— Derya Unutmaz, MD (@DeryaTR_) March 8, 2025
Other users, such as teortaxesTex, mentioned that Manus is better suited for regurgitating information like other large language models, rather than handling complex tasks. Another user, TheXeophon, noted that the Manus agent failed to mention the Nintendo Switch during its research of the gaming console market.
Moreover, a viral video showcasing the Manus AI agent automating 50 tasks was later revealed to be misleading. Yichao “Peak” Ji, Manus’s chief scientist, confirmed, “this video is definitely NOT Manus,” complete with a laughing emoji.
Despite these early missteps, it’s crucial to keep in mind that Manus AI is still in its closed beta phase, making it too soon to dismiss its potential. However, we must also approach new AI products with a level of skepticism. While Manus may not be revolutionary, it represents an ambitious step in the right direction.
As AI models become increasingly capable in task-oriented scenarios, products built upon these models will inevitably improve. The Manus AI team has already mentioned plans for significant enhancements before its public release. The question remains whether it can fulfill its promise, but it’s certainly a development worth watching.
