AI startup Butterfly Effect introduced Manus on March 6, 2025. Manus positions itself as a general AI agent that acts autonomously. Manus is getting a lot of attention leading up to general availability and is worth watching.
Manus demonstrates the potential of agentic AI
Company highlights
Butterfly Effect, a startup company, was founded in December 2022 by Red Xiao Hong.
- On March 6, 2025, Manus was introduced.
- On March 11, Butterfly Effect and Alibaba announced a partnership to develop a Chinese version of the application and begin collaboration on open-source AI models.
Manus highlights
- Manus is a next-generation AI assistant, positioned as the world's first autonomous general agent, capable of performing as a self-sufficient digital worker without requiring continuous inputs from humans.
- Manus was built using existing large language models (LLMs), including Claude from Anthropic and fine-tuned versions of Qwen from Alibaba.
- Manus appears to include AI agent capabilities of OpenAI's Operator and research capabilities of OpenAI's Deep Research.
- At this time, Manus has not been released as a generally available product, but the product is available for Research Preview, invitation only, to interested parties. According to Manus AI, within seven days of Manus' announcement on March 6, over 2 million people are waitlisted to receive an invitation to access the code ("7 Days," Manus AI, 2025).
Manus AI agentic capabilities
Manus AI sets the benchmark for traditional digital assistants by delivering end-to-end autonomous task execution. It can address complex multi-step tasks independently. Key capabilities include:
Autonomous task execution
Manus AI can perform a wide range of tasks independently, including:
- Creating comprehensive reports.
- Creating and manipulating spreadsheet data.
- Performing in-depth data analysis.
- Produce a variety of content formats.
- Plan detailed travel itineraries.
- Process tasks asynchronously, allowing tasks to continue uninterrupted and without human intervention.
Multimodal features
Manus AI can ingest and produce multiple types of data, including:
- Text (e.g. producing reports, responding to queries).
- Images (e.g. assessing and evaluating images).
- Code (e.g. augmenting and automating coding tasks).
Use of other tools
Manus AI leads the industry with its ability to leverage, integrate, and orchestrate other agents and external tools. This includes:
- Using web browsers to access and process real-time information.
- Using code editors to provide AI-assisted programming for developers.
- Leveraging relational databases to manage structured data.
Adaptive learning and optimization
Manus AI has the ability to adapt and learn from its interactions with the user to plan and optimize its task list to deliver personalized and efficient responses.
Source: Hugging Face
Manus use cases
Manus use case gallery examples
- Amazon Store Operation Analysis: Analyzing sales data, generating visualizations, and creating strategies to improve sales performance. This involves processing large data sets and providing actionable insights.
- Financial Analysis (e.g. Tesla stock): Diving deep into stock data, compiling insights, and potentially forecasting trends. This requires accessing and processing financial data, conducting calculations, and interpreting results.
- Market Research (e.g. AI products for clothing industry): Conducting in-depth research and analyzing products and competitive positioning. This involves information retrieval, comparison, and synthesis of data.
- Supplier Sourcing: Identifying suitable suppliers based on specific requirements, which includes researching companies, evaluating criteria, and potentially negotiating.
- Travel Planning (e.g. trip to Japan): Generating personalized itineraries, including flights, accommodation, and activities. This involves accessing travel information, making choices based on preferences, and creating a cohesive plan.
- Candidate Interview Scheduling: Organizing interviews for multiple candidates with optimal time management. This requires scheduling, coordination, and optimization.
- Creating Interactive Courses (e.g. momentum theorem): Designing educational content with interactive elements. This involves structuring information, developing engaging materials, and potentially incorporating multimedia.
- Developing Video Presentations: Creating engaging video content to explain complex topics. This involves content creation, visualization, and multimedia production.
Demonstration of B2B supplier sourcing from Manus' use case library. From the Manus website: "Learn how Manus handles real-world tasks through step-by-step replays."
Source: "Best Price," Manus, 2025
The introduction of Manus has captured worldwide interest
What is Manus AI and is it having a DeepSeek moment?
Euronews, March 11, 2025
Everyone in AI is talking about Manus. We put it to the test.
MIT Technology Review, March 11, 2025
After DeepSeek: China's Manus – the hot new AI under the spotlight
Asia Times, March 12, 2025
China's Manus AI partners with Alibaba's Qwen team in expansion bid
Reuters, March 12, 2025
Manus AI invite codes selling for a fortune on social media
The Economic Times, March 12, 2025
China's Manus follows DeepSeek in challenging US AI Lead
Bloomberg, March 10, 2025
China's autonomous agent, Manus, changes everything
Forbes, March 8, 2025
China's new agent Manus sparks hype – and skepticism
Business Insider, March 10, 2025
Chinese startup says it's built 'world's first' fully autonomous AI
The Independent, March 10, 2025
Manus mania is here: Chinese 'general agent' is this week's 'future of AI' and OpenAI-killer
The Register, March 10, 2025
China's new AI model 'Manus' creates global buzz, challenges OpenAI, Google
Business Standard, March 10, 2025
With Manus, AI experimentation has burst into the open
The Economist, March 13, 2025
'Outperforming DeepResearch': New Chinese AI agent Manus rewrites autonomy rulebook. What sets it apart?
Business Today, March 9, 2025
Manus Generative AI Assistant (GAIA) performance benchmarks beat OpenAI
The benchmark levels, as defined by Mialon et al. in "GAIA: A Benchmark for General AI Assistants," are as follows:
- Level 1 questions generally require no tools, or at most one tool but no more than 5 steps.
- Level 2 questions generally involve more steps, roughly between 5 and 10 and combining different tools is needed.
- Level 3 are questions for a near perfect general assistant, requiring to take arbitrarily long sequences of actions, use any number of tools, and access to the world in general.
Source: Manus
Note: The performance benchmarks above need to be validated by a third party.
Manus Generative AI Assistant (GAIA) benchmarks vs. other AI models
|
Model |
GAIA Benchmark Accuracy (%) |
Release Date |
Key Features |
|---|---|---|---|
|
Manus AI |
>65% (Assumed State of the Art) |
March 2025 (Est.) |
Autonomous execution, multi-modal, tool integration |
|
H2O.ai (h2oGPTe) |
65% |
December 2024 |
Enterprise-grade AI, tool-enhanced performance |
|
Google (Langfun) |
49% |
July 2024 |
Advanced reasoning, limited external tool use |
|
Microsoft (o1) |
38% |
2024 |
Open AI model with moderate capabilities |
|
OpenAI (GPT-4o) |
32% |
August 2024 |
Plugin-based functionality |
|
OpenAI (GPT-4 Plugins) |
15%-30% |
2023 |
Early iteration with limited real-world performance |
Source: Hugging Face
Note: The performance benchmarks above need to be validated by a third party.
Manus differentiators
- All-in-One OpenAI Operator and Deep Research Offering
- Manus is execution focused, it completes tasks without extensive human intervention, and it functions as an active autonomous AI agent rather than a passive recommendation engine
- Transparency
- Manus provides a level of transparency that developers can leverage to automate workflows. Developers can monitor how it performs real-time web browsing, API integration, and data gathering to perform task automation.
- Autonomous Task Execution
- Manus has been designed to be autonomous and capable of executing multi-step tasks with minimal human input. This digital assistant is capable of taking high-level goals and decomposing them into smaller tasks that can be easily executed.
- Multi-Agent Architecture
- Manus leverages its multi-agent architecture to perform a variety of diverse tasks efficiently and concurrently to optimize its performance for complex workflows. This enables it to manage multiple screens and concurrent tasks and makes it ideal for complex workloads.
- Continuous Learning
- Manus' capabilities for continuous learning allow it to adapt and personalize to user behavior, improving accuracy and goal alignment for developers building agentic AI applications.
Action items to consider for your organization
- Manus has to the potential to be transformative and influence agentic AI initiatives across all industries.
- Be prepared to compare your current agentic AI initiatives with what is possible.
- Be prepared to reassess your current plans with Open AI's Operator and Deep Research once the Manus product and pricing is made available.
- Prepare for more innovations, especially regarding AI software methodologies.
- Days after the Manus announcement, numerous open-source alternatives were made available.
Related Info-Tech research and other resources
Prepare to Negotiate Your Generative AI Vendor Contract
Build Your AI Solution Selection Criteria
Adopt a Structured Acquisition Process to Ensure Excellence in Gen AI Outcomes
Assessing DeepSeek: Disruption in the AI Industry
Manus Resources
Works cited
LLMHacker. "Manus AI: The Best Autonomous AI Agent Redefining Automation and Productivity." Hugging Face, 6 March 2025. Web.
Manus. "Best Price for Rubber Mats." Manus, 2025. Web.
Manus. Manus AI, 2025. manus.im
Manus AI. "7 days. 2 million people on the waitlist. We're so excited and humbled by the incredible demand, and we're working around the clock to bring Manus to more of you as soon as possible." LinkedIn, 13 March, 2025. https://www.linkedin.com/posts/manus-im_7-days-2-million-people-on-the-waitlist-activity-7305624156048351232-OIFN
Mialon, Grégoire, et al. "GAIA: A Benchmark for General AI Assistants." International Conference on Learning Representations, 2024. Web.
Olteanu, Alex. "Manus AI: Features, Architecture, Access, Early Issues, and More." DataCamp, 10 March 2025. Web.
