With a bang, the Chinese AI startup Butterfly Effect has released its highly acclaimed AI agent Manus to the general public. What was previously reserved for a small, exclusive beta community is now open to everyone: Manus, touted as an "autonomous generalist" among AI agents, can now be used without invitation or waiting list – including 1,000 free credits and a daily free task. This opening not only marks a paradigm shift in access to highly developed AI agents, but also raises questions about new possibilities, limitations, and the use of artificial intelligence in everyday life. While social media channels are literally exploding with enthusiasm and initial reports, critical assessments are inevitable: What can Manus really do, how fair is the free model, and what does this mean for the trade in beta accounts, which had recently created a flourishing gray market? In the following, we delve deep into the world of Manus, classify it, compare it with the competition and take a look at the future of AI agents.
The Unleashing: Manus becomes accessible to all
From the beta elite to a mass phenomenon
Just a few weeks ago, Manus was a myth, a coveted commodity in the AI community. Access was only granted to those who secured an invitation – a system that evoked memories of the early days of Clubhouse or Gmail. The demand was enormous: the official Discord channel boasted over 186,000 members, but less than one percent of the waiting list actually gained access. Beta accounts circulated in forums and on X (formerly Twitter), sometimes for three-digit prices. With its launch on May 13, 2025, this chapter is history. Manus is now freely available to everyone, with easy access via the web, iOS, or Android app.
The decision to open Manus was no accident. After a successful $75 million funding round led by Benchmark Capital and a $500 million valuation, the pressure to scale was as intense as the competition from global tech giants. Opening up is thus also a signal to the market: Manus aims to establish itself as the standard for autonomous AI agents before others close the gap.
Social Buzz: Enthusiasm and Skepticism
The reactions on social media were quick to arrive. "Finally, no more waiting – Manus is live!" one user on X rejoiced, while another warned: "1,000 credits go by quickly, and then it gets expensive." The initial testimonials read like a mixture of amazement and pragmatism: "Manus feels like a hyper-intelligent intern who doesn't sleep," wrote one user, "but sometimes he misunderstands tasks or delivers unusable results." The community primarily celebrates the system's autonomy and versatility, but remains critical of its actual suitability for everyday use and the pricing model.
What makes Manus special: technology, architecture and application spectrum
Autonomy as a guiding principle
At the heart of Manus is a promise: The AI agent should not only respond to prompts, but also be able to think, plan, and act independently. While traditional chatbots like ChatGPT or Gemini rely on interaction, Manus works asynchronously and autonomously in the background. Users formulate a goal—such as analyzing stock markets or planning a trip—and Manus handles all the research, planning, and execution. The results are provided in various formats (e.g., CSV, Excel, web pages).
Manus's autonomy is based on a multi-agent architecture. Various specialized sub-agents perform tasks such as research, planning, code generation, or data analysis and orchestrate complex workflows in parallel. This enables Manus to handle tasks with a depth and breadth that goes beyond traditional AI assistants.
Technological foundation: models and benchmarks
A key difference from many Western AI agents is the diversity of models. Manus combines several AI models, including Anthropic's Claude 3.5 Sonnet, fine-tuned versions of Alibaba's open-source Qwen, and other proprietary agents. This model fusion enables greater flexibility and adaptability to different tasks.
In independent benchmarks such as the GAIA test—a collaboration between Meta AI, Hugging Face, and AutoGPT—Manus performed better than OpenAI's Deep Research in several disciplines, particularly in the depth of research and the ability to structure and process complex tasks.
Application spectrum: From analysis to automation
Manus's applications are diverse and range from data analysis and the automation of repetitive tasks to the planning of complex workflows. For example, when analyzing application documents, Manus can not only evaluate and assess resumes, but also consider individual criteria that can be adjusted throughout the process. In the financial world, Manus can analyze market data, identify trends, and create interactive dashboards. The agent also excels in coding: Manus not only writes code snippets, but also develops complete programs, integrates APIs, and performs independent testing.
The ability to perform tasks in the background while the user is busy with other things sets Manus apart from traditional AI chatbots. The system provides updates on progress and results without the need for constant follow-up or correction.
Limits and challenges
Despite all the enthusiasm, there are also clear limitations. Manus is not infallible: In tests, the agent sometimes misunderstood tasks or delivered results that did not meet expectations. The context length—the amount of information Manus can process in one pass—is also limited and can quickly be exhausted for very complex tasks. Furthermore, the quality of the results depends heavily on the formulation of the task. Those who ask imprecise questions may receive imprecise or irrelevant answers.
Another issue is transparency: While Manus provides detailed reports, it often remains unclear which sources and methods the results are based on. This is a risk that should not be underestimated, especially for sensitive or security-relevant tasks.
The end of the beta grey market: From exclusive access to the mass product
The grey market for beta accounts
The exclusivity of the Manus beta had spawned a thriving gray market. Beta accounts were traded in relevant forums and on Discord, sometimes for three-digit sums. Demand was enormous, but access codes were scarce. For many, owning a Manus account was a status symbol, a ticket to the future of AI.
With the opening, this market collapsed virtually overnight. Access codes became worthless, and trading beta accounts became obsolete. While this may be annoying for some early adopters, it's a win-win for the wider community: Access to sophisticated AI has been democratized, and the barriers to entry have fallen.
Fairness and Free Credits: The new pricing model
With this launch, Manus is adopting a freemium model that sets it apart from many competitors. Every new user receives 1,000 free credits as a welcome gift, plus 300 credits per day that can be used for one free task per day. This allows even more complex projects to be tested without financial hurdles. Only those who want more—such as multiple parallel tasks, particularly long or computationally intensive projects—will have to upgrade to a paid subscription.
The "Starter" subscription costs around €43 per month (or $39) and includes 3,900 credits that can be rolled over to the following month. Power users can access further upgrades and purchase additional credits. The model is transparent and fair—at least compared to many other AI offerings, which have massive limitations in their free versions.
Criticisms of the credit system
However, there is also criticism: Credits are quickly used up, especially for complex tasks. A single deep research task can consume several hundred credits. Anyone who wants to use Manus intensively will hardly be able to avoid a subscription. Furthermore, it is still unclear how the pricing and credit system will develop in the long term – there remains concern that the barrier to entry could rise as the number of users grows.
Competition and alternatives: The race for the best AI agent
OpenAI Operator: The US giant as a challenger
OpenAI has its own AI agent, "Operator," which, similar to Manus, can autonomously perform tasks on the web—from forms to online orders to appointment scheduling. However, Operator is currently only available to Pro subscribers in the US and doesn't yet achieve human-level accuracy in benchmarks such as OSWorld and WebArena. Availability is also very limited, and a global rollout is still pending.
Qwen Deep Research: Open Source from China
With Qwen, Alibaba has established an open-source model that also handles complex research tasks with its new "Deep Research" feature. Qwen is completely free, offers comparable research depth, and is primarily aimed at users who value transparency and customizability. The integration of multi-step planning, web search, and report generation makes Qwen a viable alternative, especially for tech-savvy users.
Deepseek: The pioneer from the Far East
Deepseek, also from China, caused a stir earlier this year because its system set new standards in terms of autonomy and versatility. However, Deepseek is severely limited when it comes to politically or socially sensitive topics—a problem that is also being discussed with Manus, although initial tests show that Manus provides more detailed responses than Deepseek on critical topics such as the Tiananmen protests.
Comparison table: The most important AI agents at a glance
| system | Access | Cost | autonomy | Special features | restrictions |
|---|---|---|---|---|---|
| Manus | Open | 1,000 credits + 300/day free | High | Multi-agent, multiple models | Credits used up quickly |
| OpenAI Operator | US Pro-Only | Pro subscription | Medium | Web interaction, OpenAI models | USA only, limited |
| Qwen | Open | For free | High | Open Source, Deep Research | Technically demanding |
| Deepseek | Open | For free | Medium | High speed | Political filters |
This overview shows that Manus is currently leading the way with its open access, multi-agent architecture, and model diversity, but also faces challenges such as the credit system and result quality.
Possibilities and limitations: What Manus cannot (yet) do today
Strengths: Autonomy, versatility, productivity
Manus's greatest strengths lie in its autonomy and ability to perform complex tasks without constant user interaction. Its multi-agent architecture ensures efficient task sharing, and its model diversity ensures flexibility and adaptability.
Manus can analyze data, create reports, write code, automate workflows, and even generate custom websites. The system opens up new dimensions of productivity, especially for developers, analysts, and knowledge workers.
Weaknesses: Context, transparency, susceptibility to errors
But Manus isn't a jack-of-all-trades. The context length is limited, and the system reaches its limits with very complex tasks. The results aren't always comprehensible or transparent, which can be problematic, especially with sensitive topics. Furthermore, the susceptibility to errors – for example, with unclear task definitions or ambiguous prompts – should not be underestimated. Manus can misunderstand tasks or deliver irrelevant results.
Another problem is the dependence on external sources and APIs. If these are unavailable or change, Manus can no longer perform tasks correctly. Data protection and security are also open questions: How is the data processed, stored, and protected? So far, there is little reliable information from the provider on this matter.
Upgrades and monetization: Between free model and power user plans
The credit system in detail
The heart of the monetization model is the credit system. Each user receives 1,000 credits at the start, plus 300 credits daily, which can be used for one task per day. Complex tasks—such as deep research or programming projects—consume several hundred credits per run. Those who want more must upgrade to a paid subscription.
The "Starter" subscription costs around €43 per month and includes 3,900 credits that can be rolled over to the following month. For frequent users, there are further upgrades and the option to purchase additional credits. Prices for add-on packages have not yet been publicly announced.
Fairness and criticism
The model is fairer than many competitors, as entry is free and the usage options in the free model are comparatively extensive. However, there remains concern that the barrier to entry could rise with growing user numbers, leading to higher prices in the long term. The question of how credits are calculated for particularly computationally intensive tasks has not yet been conclusively resolved.
Upgrades and additional features
Paid upgrades not only provide users with more credits, but also access to advanced features such as high-performance mode, editing prioritization, and improved stability. For businesses and developers, there are special API access and integration options that allow for deeper integration into existing workflows.
Outlook: The future of AI agents and the role of Manus
The race for autonomy
The launch of Manus marks a turning point in the race for the best AI agents. While traditional chatbots like ChatGPT or Gemini continue to focus on interaction, systems like Manus, Qwen, and Operator are driving the development of autonomous agents. The trend is clearly toward increased productivity, automation, and integration into everyday life.
The competition isn't sleeping: OpenAI is working on the global expansion of Operator, Alibaba is investing heavily in Qwen and Deep Research, and numerous startups around the world are developing their own agent architectures. The market will continue to fragment, but also specialize, in the coming months.
Challenges: Regulation, transparency, ethics
As autonomous AI agents become more widespread, so do the challenges. Regulation, data protection, and transparency are becoming key issues. The question of how AI agents make decisions, what data they use, and how they handle sensitive information will shape the debate in the coming years.
The social impact should not be underestimated either: Autonomous AI agents could radically change work processes, but also create new forms of dependency and control. Democratizing access—as Manus has now accomplished—is an important step, but it also carries risks, such as misuse or misapplication.
Manus in 2025: Between hype and reality
Manus exemplifies the ambivalence of current AI development: On the one hand, a quantum leap in terms of autonomy, productivity, and accessibility; on the other, unresolved problems with transparency, error susceptibility, and monetization. Opening it to everyone is a milestone, but not an end point. Manus must now prove itself in everyday life, win over the community, and overcome the challenges of scaling.
The coming months will show whether Manus can turn the hype into sustained success—or whether the competition will become faster, cheaper, and better. One thing is certain: the race for the best AI agent has only just begun.
Conclusion
With the opening of Manus to all users, a new chapter in the history of AI agents has begun. The system impresses with its autonomy, versatility, and a fair freemium model, but also encounters limitations in terms of context, transparency, and pricing. The gray market for beta accounts is history, and access to sophisticated AI is democratized.
The competition isn't sleeping: Qwen, Operator, and Deepseek offer strong alternatives. The future of AI agents remains exciting – Manus has set the bar high, but it has to assert itself in a dynamic environment. Anyone who wants to experiment with AI today can't ignore Manus – whether they're a curious beginner, an ambitious developer, or a critical observer.
Further links:
- After Deepseek comes Manus: We tried out the AI agent from China
- China's Manus AI offers free registration after fresh funding amid AI agent competition
- Manus AI public, free credits
- Manus: The much-hyped Chinese AI has opened up public access and you get 1,000 credits for free if you sign up now
- Introducing Manus: The General AI Agent
- AI agent Manus introduces subscription fees of up to 220 euros
- OpenAI Operator
- AI Base News