Manus, autonomous AI agent launched in preview last week, announced on Tuesday that it had formed a strategic partnership with Alibaba’s large model Tongyi Qianwen.
The Chinese version of Manus will fully function on domestic AI models and computing platforms, based on Tongyi Qianwen’s open-source model and in collaboration with Alibaba’s AI team, the company said.
“We look forward to bringing Manus’ innovative experience to a broad Chinese-speaking audience as soon as possible. Stay tuned,” Manus said on its website, though it did not disclose a specific launch date for the Chinese version.
Manus is a project incubated by the startup Monica.im, which was founded in July 2023. Monica.im specializes in AI browser plugins and tools for consumers, utilizing third-party AI models. Its founding team includes serial entrepreneurs Xiao Hong and Ji Yichao, along with Zhang Tao, who previously worked at ByteDance as a product lead.
Monica.im’s products were initially targeted at overseas markets. While it generated some revenue through consumer subscriptions, it did not gain widespread recognition outside the AI community due to the saturation of AI applications and chatbot products.
On March 6, Monica.im released an early preview of its AI agent tool Manus, claiming it to be the first general AI agent capable of solving complex tasks like a human. In a demonstration video, Ji Yichao showcased Manus’ capabilities, including automated resume screening, real estate market research, and stock analysis.
Unlike foundation models such as DeepSeek-R1 and GPT-4, Manus is not a standalone large language model but rather an AI product that calls upon third-party foundation models. Similar open-source projects exist in the market. Ji Yichao revealed that Manus currently uses models from the U.S.-based AI company Claude and a fine-tuned version of Alibaba’s Qianwen model. Compared to chat-focused AI applications like ChatGPT and DeepSeek, Manus aims to specialize in understanding and executing complex instructions, bringing AI processing closer to real-world task execution.
Following its overseas release on March 6, the limited availability of early preview access codes led to significant hype on social media. Some self-media outlets fueled speculation, and at one point, Manus’ “beta access codes” were resold on secondary markets for tens of thousands of yuan.
Discussions and reviews about Manus became one of the top trending AI-related topics across content platforms, second only to DeepSeek. However, current user evaluations suggest that Manus still requires human intervention to complete tasks effectively, indicating that it is not yet a fully realized product.
On March 6, Zhang Tao commented on social media that Manus remains a demo with considerable room for improvement before becoming a fully developed product. The invitation-only system was primarily due to server capacity limitations. “The server resources were allocated based on typical demo release standards—we never expected such a huge response,” Zhang wrote.
Details regarding Manus’ collaboration with Alibaba remain unclear. Alibaba is one of China’s largest investors in AI. On February 24, the company announced plans to invest over 380 billion yuan in the next three years to build cloud and AI infrastructure, surpassing its total capital expenditure over the past decade. Alibaba stated that this investment is the largest AI and cloud infrastructure investment ever made by a Chinese private enterprise.
On March 6, Alibaba also launched QwQ-32B, a new inference model designed to rival OpenAI’s o1 model. Alibaba claimed that QwQ-32B outperformed OpenAI’s o1-mini in benchmark tests and matched DeepSeek-R1 in evaluation capabilities. Despite having only 32 billion parameters, the model significantly reduces deployment requirements, making local deployment possible even on consumer-grade GPUs, according to Alibaba.