Manus AI agent is China’s latest artificial intelligence breakthrough that’s turning heads in Silicon Valley and beyond. Manus was launched last week via an invitation-only preview, and represents China’s most ambitious entry into the emerging AI agent market.
Unlike anything seen to date, the Manus AI agent doesn’t just chat with users – it is allegedly capable of independently tackling complex multi-step tasks with minimal human guidance.
Developed by Chinese startup Butterfly Effect with financial backing from tech giant Tencent Holdings, Manus AI agent has captured global attention for its ability to bridge the gap between theoretical AI capabilities and practical, real-world applications. It uses an innovative multi-model architecture that combines the strengths of multiple leading language models.
Breakthrough autonomous task execution
In a post on X, Peak Ji Yichao, co-founder and chief scientist at Butterfly Effect, said that the agentic AI was built using existing large language models, including Anthropic’s Claude and fine-tuned versions of Alibaba’s open-source Qwen.
Its multi-model nature allows Manus to use different AI strengths according to what’s demanded of it, resulting in more sophisticated reasoning and execution capabilities.
“The Manus AI agent represents a fundamentally different approach to artificial intelligence,” CNN Business stated. According to coverage, Manus “can carry out complex, multi-step tasks like screening resumés and creating a website,” and “doesn’t only generate ideas but delivers tangible results, like producing a report recommending properties to buy based on specific criteria.”
Real-world performance assessment
In an extensive hands-on evaluation, MIT Technology Review tested the Manus AI agent in three distinct task categories: compiling comprehensive journalist lists, conducting real estate searches with complex parameters, and identifying candidates for its prestigious Innovators Under 35 program.
“Using Manus feels like collaborating with a highly intelligent and efficient intern,” wrote Caiwei Chen in the assessment. “While it occasionally lacks understanding of what it’s being asked to do, makes incorrect assumptions, or cuts corners to expedite tasks, it explains its reasoning clearly, is remarkably adaptable, and can improve substantially when provided with detailed instructions or feedback.”
The evaluation revealed one of the Manus AI agent’s most distinctive features – its “Manus’s Computer” interface, which provides unprecedented transparency into the AI’s decision-making process.
The application window lets users observe the agent’s actions in real time and intervene when necessary, creating a collaborative human-AI workflow that maintains user control while automating complex processes.
Technical implementation challenges
Despite impressive capabilities, the Manus AI agent faces significant technical hurdles in its current implementation.MIT Technology Reviewdocumented frequent system crashes and timeout errors during extended use.
The platform displayed error messages, citing “high service load,” suggesting that computational infrastructure remains a limitation.
The technical constraints have contributed to highly restricted access, with less than 1% of wait-listed users receiving invite codes – the official Manus Discord channel has already accumulated over 186,000 members.
According to reporting from Chinese technology publication36Kr, the Manus AI agent’s operational costs remain relatively competitive at approximately $2 per task.
Strategic partnership with Alibaba Cloud
The creators of the Manus AI agent have announced a partnership with Alibaba’s cloud computing division. According to a South China Morning Post report dated March 11, “Manus will engage in strategic cooperation with Alibaba’s Qwen team to meet the needs of Chinese users.”
The partnership aims to make Manus available on “domestic models and computing platforms,” although implementation timelines remain unspecified.
Parallel advancements in foundation models
The Manus-Alibaba partnership coincides with Alibaba’s advances in AI foundation model technology. On March 6, the company published its QwQ-32B reasoning model, claiming performance characteristics that surpass OpenAI’s o1-mini and rivalling DeepSeek’s R1 model, despite a lower parameter count.
CNN Businessreported, “Alibaba touted its new model, QwQ-32B, in an online statement as delivering exceptional performance, almost entirely surpassing OpenAI-o1-mini and rivalling the strongest open-source reasoning model, DeepSeek-R1.”
The claimed efficiency gains are particularly noteworthy – Alibaba says QwQ-32B achieves competitive performance with just 32 billion parameters, compared to the 671 billion parameters in DeepSeek’s R1 model. The reduced model size suggests substantially lower computational requirements for training and inference with advanced reasoning capabilities.
China’s strategic AI investments
The Manus AI agent and Alibaba’s model advancements reflect China’s broader strategic emphasis on artificial intelligence development. The Chinese government has pledged explicit support for “emerging industries and industries of the future,” with artificial intelligence receiving particular focus alongside quantum computing and robotics.
Alibaba will invest 380 billion yuan (approximately $52.4 billion) in AI and cloud computing infrastructure in the next three years, a figure the company notes exceeds its total investments in these sectors during the previous decade.
As MIT Technology Review’s Caiwei Chen said, “Chinese AI companies are not just following in the footsteps of their Western counterparts. Rather than just innovating on base models, they are actively shaping the adoption of autonomous AI agents in their way.”
The Manus AI agent also exemplifies how China’s artificial intelligence ecosystem has evolved beyond merely replicating Western advances. Government policies promoting technological self-reliance, substantial funding initiatives, and a growing pipeline of specialised AI talent from Chinese universities have created conditions for original innovation.
Rather than a single approach to artificial intelligence, we are witnessing diverse implementation philosophies likely resulting in complementary systems optimised for different uses and cultural contexts.