Image source: The speed of Manus breaking the circle generated by unbounded AI has also set records set by ChatGPT and DeepSeek, and it has become popular on the Internet within one day of internal testing!

For this company in Wuhan called "Butterfly Effect", it is actually the second time that they have made a hit product. The first one is Monica, an AI product with 10 million users overseas. Its annual recurring revenue ARR reaches tens of millions of US dollars, making it one of the most revenue-capable AI products in 2024.

Manus is the team's first AI Agent product. During the internal testing period on March 6, it was rated by many industry insiders as a product that "started the first year of Agent" by many industry insiders. From its popularity to today, it has only been 4 days since it was released, and there has not even been a public beta. With a few cases, everyone has hit this product with the altar, and the internal test code is up to 50,000 yuan.

In fact, the company was acquired by Bytes for $30 million in 2024. Although they ended up feeling that the price was not fair enough, the acquisition was not completed. Brother Jing exclusively learned from multiple sources that the inside story is that Byte believes that Monica products are more "shell-in-covered", the product investment cost is very high, and the data retained is not optimistic, and it may be replaced by a big model in the future. He will also launch a bean bag plug-in, so the quote given is a bit lower than the price.

For Monica, the acquisition of ARR companies with 10 million US dollars is only 3 times the price-to-sales ratio concept (using the secondary market price-to-sales ratio concept) is several million higher than their valuation in the previous round, which is obviously not in line with the expectations of investors behind it.

In the era of mobile Internet, companies and products were sold to a unicorn. After the AI ​​era began, from Monica to Manus, the new company's product strength and execution ability are quite strong. Sequoia China and Tencent in the second round decided to support their further entrepreneurship. After all, in the field of AI native, where big companies lack popular products, two products have been made with only a small amount of financing.

In particular, Manus's popularity is accompanied by amazing praise. Many people think that it is a DeepSeek-level innovation, the key to opening the door of AI Agent, and even a representative product to the road to AGI. At the same time, it is also accompanied by the sarcasm of the tide. Many people think that without core technology, it is just a shell product; some people conspire to be a joint KOL marketing, and no splash abroad has caused any splashes; some people think that the subsequent opening service cost is too high and it may be difficult to implement.

After many different opinions, Brother Whale also studied this product in depth, trying to share with you the breakthrough and significance of this product from a more objective and long-term perspective.

DeepSeek breaks the calculation power standard, Manus starts the first year of Agent

In MarchOn the 6th, at a small press conference on Manus products. The company put up a slogan "The next ChatGPT moment", describing the door to the Agent era has been opened.

Whether Manus is enough for the next GPT moment, it is still impossible to judge. DeepSeek, which has been popular for a month, has proved its strength.

DeepSeek became popular during the Spring Festival because everyone experienced the charm of reasoning models for the first time. It's really smart. Many of the answers are in depth, breadth and flexibility, which exceed the well-known products such as bean bread and ingots. This is the advantage of inference models over instructional models, and it is also the intelligence of DeepSeek.

In 2025, big models face three breakthrough directions: large parameters, multimodal, and reasoning. The first path is that Grok 3 is still strongly through 200,000 H100 graphics cards, hard breaking through the Scaling law of the big model, while other companies are becoming bigger and stronger through architectures such as MoE; multimodal is the main direction for everyone to focus on. OpneAI abroad, domestic Byte Beanbao, Tencent Hunyuan, Step Yuexing, etc. are all representatives, integrating DIT architecture, and being proficient in everything you hear and write; the third path is to improve the model cleverness through reinforcement learning RL and other methods. DeepSeek is the representative of this direction, and Tongyi will follow up quickly.

DeepSeek's actually has a bigger feature, which is its excellent cost control capability, especially projects that have been open sourced for 5 consecutive days since then. This ability to optimize underlying facilities has also gradually confirmed, breaking the technology and financial order created by Nvidia and other companies.

What is the "computing power-based"? The United States' financial hegemony has undergone three underlying logical leaps: from gold, oil to computing power, which is essentially reshaping the credit system by monopolizing global core resources and maintaining the hegemony of the US dollar.

The collapse of the gold standard (1944-1971) The Bretton Woods system established the linkage between the US dollar and gold, but the insufficient US gold reserves led to the collapse of the system. In 1971, Nixon announced that the US dollar was decoupled from gold, and the US dollar urgently needed a new anchor. Petro-USD hegemony (1974-now) The United States and Saudi Arabia have secretly agreed to lock in oil settlement in US dollars, establishing a "oil-USD-US bond" cycle: oil exporters earn US dollars and purchase US bonds, and the US dollar becomes the global reserve currency. At its peak, 86% of global oil trade was settled in US dollars, and the Federal Reserve reaped global wealth through US dollar tides. The rise of computing power is the first priority (2020s-) Computing power has become a new means of production in the digital age. Nvidia H100 chips have become "computing power currency". In 2023, the global computing power market size reached US$2.6 trillion, and American companies account for 60%. Computing power is replacing oil and becoming the new pillar of US dollar credit.

The amazing computing power demand of the big model, the computing power standard behind it, has led to the cumulative increase of Nvidia's stock price in the past two years, with a market value of more than 30 in the past 10 years.The $0 billion rose to nearly $3 trillion.

For Manus, it has indeed not caused any shock in the foreign technology circle. It cannot affect Nvidia's stock price fluctuations like DeepSeek. But according to partner Zhang Tao, the company has only more than 50 people, and it took two or three months to create this popular AI product.

Comparison of WeChat search index: Manus has not surpassed its peak Deepseek

But in the domestic market, Manus is still the hottest AI product. Just by releasing a few examples of the website, it can set off a storm of discussion on the Internet. As one code is hard to find, many people even ask for code at a high price. An application of Maunus invitation code generator even topped the iOS domestic payment list on March 8.

Of course, this product is useless, but it has gained traffic. Manus has a leading insight and lacks innovation in major manufacturers

For most people, they have never experienced OpenAI Deep Research (up to $200 per month), the domestic Manus is indeed a bit amazing after getting started. Brother Whale tested the following questions:

A problem like making an embossed intelligent report is stuck when analyzing the data for the first time, and the link that requires the final PPT is undesirable. So the second time requires the text report, which is also stopped before making the chart.

At present, Manus often cannot understand and control his abilities, and often chew too much.

In principle, Manus is not complicated. Manus is an AI product that integrates Computer use, virtual machines, and Multi agent collaboration.

Note: From Baoyu AI

But the most important breakthrough of Manus is that compared with automated programming software Devin and bolt.new, it has achieved the productization of Agents in the general field.

Employees who have worked in this company mentioned on social media that the company's engineering practice and agent workflow have a deep accumulation:

The first agent was launched in China from September to October 23, and the todolist.md in it was the best practice after learning the agent solutions at that time. I started the GPT platform in March 2019. Since the beginning of 2019, I have been accumulating browser technology and accumulating a lot of understanding of browser context utilization. I started searching in November 2019 and have accumulated the ability of agents to obtain information online. The part I didn't participate in. In July 24, we gained experience in social communication growth through roast. Understanding of coding capabilities in coding products in November 2019. Erix

"ReallyEverything is a relatively thin layer, but it is also a fact that the combination innovation formed by these building blocks in this window is strong enough. "This employee is evaluating this.

Brother Jing believes that Manus's most successful product insight beyond the big manufacturers.

Xiao Hong, founder of Butterfly Effect, previously launched the "Yiban Plugin" as a serial entrepreneur. The "Yiban Plugin" that was previously launched was a WeChat plug-in product with a revenue of tens of millions (the new media operation of Jingxuan Account is also paid for). Since then, he has seized the SCRM trend of Qiwei and became a Microban Assistant. When the AI ​​trend comes, it has made a big model collection product Monica. Now the AI ​​Agent trend is coming and Manus has been launched again.

Especially, the success of two consecutive AI native products is not easy. It is as strong as OpenAI, in fact, in addition to ChatGPT In addition, the other products are not successful, many of them are in semi-finished state, such as GPTs, SearchGPT, DALL.E, Whisper, etc.

At present, the AI ​​products of major domestic manufacturers generally lack some creativity. From AI social networking to AI search, and AI Coding products, they are all products that people have.

This picture is generated by Tencent Yuanbao AI

But Butterfly Effect Company, after using Monica to seize the first wave of AI plug-in dividends, was the first to make the first mature product in the Agent. Among the five AI stages shared by Ultraman Sam, L1 (chatbot), L2 (reasoner), L3 (agent), and Manus successfully positioned the initial breakthrough products of the L3 stage.

Especially their "Less Structure more Intellingence" concept made them give up the AI ​​browser's path that cannot be used as a big company, and also made them discover new vitality.

As for whether this wave is the out-of-the-box effect caused by marketing leverage, Brother Whale believes that it is not the case.

Previously, Monica released a Chinese version and had a wave of cooperation with KOL, which was to issue free membership quotas through KOL. When Manus asked KOL to attend a small press conference, he did tell everyone that it was the adjective "the world's first universal Agent product". I learned that KOLs did not have a cooperative promotion.

It is estimated that if Manus did not have a big response, he would also release a free token quotas through KOL like Monica. Now it is definitely not necessary, but judging from the WeChat search index in the previous article, it is not as popular as DeepSeek.

Since Capital Soul Capital partner Wu Bingjian once described: After Deepseek broke the circle, society was indeed educated by facts, and everyone started to write a technical level - how to improve Attention, how to improve MoE, how to mix FP8 and FP16. Going further, it is the original level - who can innovate the next generation model architecture, and who can discover the next set of training modelsmethod.

So from this perspective, Manus also brings everyone's product strategy back to the field of cutting-edge innovation, focusing on what kind of innovative products can everyone make after the first year of AI application/Agent, rather than the investment war of AI assistants.

The shell does not affect it, but it is still far from success.

At present stage AI products do not essentially have core barriers and competitiveness. It is difficult to register as a patent for product concepts, and engineering capabilities are the best things for big manufacturers, so Brother Jing mentioned on Xiaohongshu before, guess which big manufacturer makes similar products.

But to be honest, Manus is an Agent product with more complete engineering delivery compared to some of its previous general agent products, such as OpenAI's Operator, Anthropic's Claude use, and Tencent's APPAgent.

However, perfection is not a very high barrier to a product. After Manus became popular, the MetaGPT team spent 3 hours developing OpenManus and open source.

After Manus explored actual needs, optimized technical engineering paths, and continuously improved product functional details, it is indeed not difficult for the outside world to copy a homework "half-open" book. It was replicated in half a day, so would Manus have no barriers?

A certain investor lost the opportunity to invest in Perplexity because of shell theory

Now countless AI search products have not affected the development of Perplexity. Native product understanding will help Perplexity continuously update better functional points, while other similar products can only be followed up and copied later.

The same is true for Manus. What may be most needed at this stage is to accept financing from a large manufacturer. After all, the test is not publicly tested, because the server's ability to accept is insufficient.

The message posted at the media communication meeting showed that the Manus team also gave the cost of running a single task: about two dollars. The cost is already 1/10 of DeepResearch, but the cost of a single task is still as high as nearly 15 yuan. This is also the reason why Manus has adopted a small-scale distribution of invitation codes, which has also caused the internal system to crash.

Accepting financing from large manufacturers is not only about funds, but more importantly, the low-cost supply of large model APIs. According to media reports participating in the communication meeting, Manus' core calls the Claude model, plus some fine-tuning of the Tongyi open source model. Will you accept investments from Alibaba, Tencent or Byte in the future, and investments from funds + resources like Kimi?

This allows Manus to launch a subscription membership service at an annual fee of less than 1,000 yuan. Otherwise, higher than this price,It can become a toy for a few professionals.

On the other hand, it is to improve product details and service capabilities, and rapid iteration is the foundation for ensuring vitality.

It is still very long to generate an answer by Manus, and many tasks will still collapse. Manus needs to make more thoughts on productization, rather than what Zhang Tao, the head of Manus product, said: "It's really very simple, it's just that there is no secret, it's just that it believes in the power of the model."

Model is the foundation, and product details are service capabilities. For example, Claude 3.7 Sonnet's code programming capabilities have once again broken the ceiling, but Cursor can still attract everyone to pay for subscriptions with the advantages of automatic code completion.

More importantly, the MCP (Multi-Agent Communication Protocol) aggregation model is already showing high growth potential. This is also the development path that Manus should adopt in the future.

From the evolution speed of the big model, built-in agents may also be a trend. In the future, GPT 5 may be surprisingly powerful if it can achieve the integration of inference and instruction model, multimodal capabilities, and built-in Agents. Domestic large manufacturers should also be following this route. Before this, Manus needed to reach users and revenue scale.

Summary:

DeepSeek breaks the foreign model success theory in the field of big models and its extremely low cost implementation ability, making everyone believe that the mysterious power of the East has the ability to impact the global science and technology financial order. In other words, DeepSeek has made China's AI competition tickets on its own, and directly ended the era of global investors in valuation of China's assets.

For Manus, it shows the strongest form of AI native, not the same Chatbot, nor the AutoAgent like a Trojan horse, but more scenarios are useful and hopefully implemented.

To be on the high point, he may be able to become WeChat in the next AI era, but it may not affect the foreign Facebook and other products becoming more popular. Speaking of low points has opened up the public's awareness of Agent and given many startup teams the confidence to continue to create dreams.