Manus AI agent put to the test, outperforms single-system chatbots

Manus, Butterfly Effect’s new AI agent, marks a significant evolution in autonomous AI capabilities by coordinating multiple AI models to handle complex tasks. Unlike single-model chatbots, this general AI agent distributes work across independently operating systems, creating what early users describe as a highly capable digital assistant. Despite minimal access—fewer than 1% of waitlisted users have received invites—Manus has already generated substantial buzz in both Chinese and Western tech communities, with influential figures like Jack Dorsey praising its performance.

The big picture: Manus represents a different approach to AI by functioning as a multi-model agent system rather than relying on a single large language model.

• The system leverages various AI models including Anthropic‘s Claude 3.5 Sonnet and fine-tuned versions of Alibaba‘s Qwen to coordinate autonomous task completion.
• This approach differs fundamentally from single-model chatbots like DeepSeek, which primarily handle conversational interactions.

Key details: MIT Technology Review tested Manus and found it functions like “a highly intelligent and efficient intern” with strong adaptability and reasoning capabilities.

• The system occasionally misunderstands instructions, makes incorrect assumptions, or takes shortcuts, but explains its reasoning clearly.
• When provided with detailed instructions or feedback, Manus demonstrates remarkable ability to improve its performance.

Behind the numbers: Interest in Manus has exploded despite limited access, indicating significant market anticipation for autonomous AI agents.

• While the exact size of the waitlist remains undisclosed, Manus’s Discord channel has accumulated over 186,000 members.
• This surge in interest echoes the earlier excitement around DeepSeek, another Chinese AI breakthrough that surprised the industry.

Performance assessment: Testing revealed Manus excels at analytical tasks requiring extensive internet research within limited scope.

• In two of three test cases, Manus outperformed ChatGPT DeepResearch, though it required significantly more time to complete tasks.
• The system proved intuitive for both technical and non-technical users, suggesting broad potential applicability.

Why this matters: Manus demonstrates that Chinese AI companies aren’t merely following Western development patterns but are charting their own innovative path.

• The emergence of sophisticated AI agents from China highlights ongoing innovation beyond base models, particularly in autonomous agent architecture.
• This development further shifts the competitive landscape of global AI development, challenging assumptions about the direction of AI advancement.