Team of Rivals: China’s Manus employs multiple AI models for autonomous task completion

Chinese startup Butterfly Effect has thrust itself into the global AI spotlight with Manus, a multipurpose AI agent system that autonomously executes complex tasks rather than merely generating ideas. Unlike traditional AI chatbots, Manus combines multiple AI models to independently research, analyze data, generate reports, and even write and deploy code without human intervention. This development represents China’s second major AI breakthrough following DeepSeek earlier this year, potentially signaling a significant advancement in autonomous AI agent capabilities.

The big picture: Manus AI emerged suddenly on March 5, 2025, when the company described it as “the first general AI agent” capable of executing complex tasks autonomously rather than just generating ideas.

• Developed by Butterfly Effect, a Chinese company with offices in Beijing and Wuhan reportedly employing only a few dozen people, Manus has rapidly gained attention in China’s AI landscape.
• The system doesn’t rely on its own foundation model but instead integrates existing large language models, including Anthropic‘s Claude 3.5 Sonnet and fine-tuned versions of Alibaba’s Qwen models.

Why this matters: AI influencers are celebrating Manus as a potential turning point for AI agent technology, with some declaring it “China’s second DeepSeek moment.”

• Unlike DeepSeek R1, which created its own foundation model, Manus represents a different approach to innovation by integrating and orchestrating existing LLMs into an autonomous multi-agent system.
• The system’s ability to independently perform complex tasks from start to finish without human intervention represents an important evolution beyond mere text generation.

Impressive capabilities: AI newsletter founder Rowan Cheung personally tested Manus and reported remarkable results across several complex tasks.

• The system created and deployed a biography website with 100 percent accuracy and real-time data retrieval.
• It analyzed San Francisco neighborhoods to identify optimal rental locations based on multiple factors including crime rates, AI industry presence, and entrepreneurship density.
• Manus also developed a comprehensive AI course, generating eight chapters of content.

Behind the numbers: Despite generating significant buzz, very few people have actually been able to test Manus AI due to severe server shortages following its sudden rise to fame.

• Currently in private beta, Manus AI is only available via invitation codes.
• The company plans to expand access soon, but server limitations may delay a full public launch.