MiniMax: The Sweetness and Hidden Worries Under the Revenue of $70 Million | Focus Analysis

Written by Zhou Xinyu

Edited by Su Jianxun

The latest news about MiniMax is presumably known to everyone. Foreign media has reported MiniMax's expected annual revenue: 70 million US dollars. This is a powerful rebuttal to the assertion that "large model companies cannot find PMF (Product-Market Fit)."

We have learned that this figure is still conservative.

Insiders disclosed to "Intelligent Emergence" that MiniMax's actual annual revenue is likely to exceed 70 million US dollars. And the main source of income comes from MiniMax's overseas C-end AI application, Talkie.

The interface of Talkie.

This AI application for role companionship that went online in June 2023, with the understanding and generation capabilities of AI and the popularity of the AI track, quickly created a growth myth overseas:

According to the monitoring of Sensor Tower, since its release, Talkie has consistently ranked among the top three in global downloads of AI applications for companionship. The other two are the well-known unicorns in the United States, Characte.AI and Replika.

It is understood that the rapid growth of users has also brought considerable advertising revenue to Talkie. Of course, as a ToC application, Talkie's charging model for users mainly focuses on membership subscriptions and recharging to unlock rights such as "card drawing".

The membership subscription interface of Talkie. Source: Internet

The success of Talkie has triggered a wave of AI applications for role companionship in China. However, neither ByteDance's "Cat Box" nor the once popular "Pinch It" in the二次元 community can replicate Talkie's global growth rate.

The difficulty in replicating Talkie's success has also led MiniMax to a deadlock when attempting to expand its business using the same methodology.

Today, MiniMax's valuation has exceeded 20 billion yuan - in order to enhance the company's business barriers and risk resistance capabilities, it needs more Talkies.

Several industry insiders told "Intelligent Emergence" that whether it is the domestic version of Talkie promoted by MiniMax, "Xingye", or the productivity tool "Conch AI", MiniMax wants to replicate the same approach on them: adding elements of companionship and social interaction.

However, even the creator of Talkie has not been able to recreate another Talkie.

Why Talkie is Popular: Early Start, Fine Tuning

The explosion of Talkie undoubtedly has high reference value for domestic entrepreneurs who are struggling to find AI PMF (Product-Market Fit).

As a company that has only emerged in the wave of large models, how did MiniMax create a product with an annual revenue of tens of millions of dollars in just one year? After interviewing several industry insiders, "Intelligent Emergence" summarizes the answer as: timing of launch, product positioning, and rapid polishing and iteration.

Early and fast are the remarkable characteristics of MiniMax's products.

In October 2022, one month before the launch of ChatGPT, MiniMax launched the predecessor of Talkie, an AI agent product that can directly communicate, named Glow. In June 2023, after Glow's trial in the domestic market, Talkie became the first product of MiniMax to "go overseas on the first day" internally.

And at this point in time, going overseas has not yet become a consensus in the AI industry.

In order to launch the product as quickly as possible, even if the AI technical capabilities still have flaws, MiniMax is willing to launch it first and then fix it.

Around October 2022, MiniMax was still in the internal testing of a self-developed large language model with 30 billion parameters. A MiniMax front-end engineer recalled to "Intelligent Emergence" that at that time, the model had many bugs that needed to be compensated for by the front-end engineering.

He gave an example: Because users like to use "..." to express speechlessness, the model mistakenly thought that all punctuation should be repeated three times, "At that time, more Instructions had to be written. In fact, the model itself is not that smart."

But the brilliance of MiniMax lies in that, compared to many products on the market that aim to show off "AI muscles" - and this phenomenon still prevails to this day - the core of Talkie is not AI, but rather the effort is put into product positioning and user experience.

"Talkie is actually not an AI application, it is a companion game." An AI product manager of a large company evaluated to "Intelligent Emergence", "From the results, Talkie is not an application to showcase AI functions. The AI function only makes its companion attribute perform better."

The companion attribute, to a large extent, determines that Talkie will have excellent user stickiness.

"The product logic of Talkie is the same as 'raising a child' (companion doll) and 'Travel Frog'." A product manager of Talkie told "Intelligent Emergence", "The role of AI is to make the feedback given by the created character more 'understanding of you' as the user's usage time increases."

He gave an example: Once, he mentioned to the AI character he created that he wanted to travel to Iceland before the end of the year. Recently, when he talked to the AI character about his annual leave, the other party mentioned: Have you fulfilled your dream of going to Iceland?

The aforementioned AI product manager also gave a set of data: Users who have used Talkie continuously for a week have a payment rate and retention rate of about 80%.

A product that seizes the blank market with the first-mover advantage can attract the attention of the first wave of users, which also means more effort needs to be put into the product functions.

In the opinion of many insiders, compared to many AI companion products, Talkie has put more effort into engineering and UI design. This also makes Talkie, even though there are technical flaws, still a mature product that can attract users as soon as it is launched.

The mobile phone of the aforementioned AI product manager of a large company has almost downloaded all the well-known AI role-playing applications on the market, such as Character.AI and Replika. He evaluated to "Intelligent Emergence" that most companies can disassemble the engineering and functional design of Character.AI, but Talkie is difficult to replicate:

"We spent a lot of effort to study (Talkie) why a paywall is set here and an interaction is added there. Later, we found that its engineering is too fine and interlocking."

There are many products that gain the first-mover advantage, but to maintain the advantage, rapid iteration based on market and user feedback is necessary.

"Every design of Talkie is rapidly iterated based on user feedback." An employee of Talkie told "Intelligent Emergence". He remembers that the design requirements of Talkie are the most in MiniMax. At the peak, the product team can put forward more than 20 requirements, and about 5 design employees need to meet them within a week, "The intensity exceeds that of many startups I was in before."

Similarly, Talkie and its domestic version "Xingye" are also the products to which MiniMax allocates the most manpower. The aforementioned employee disclosed that there are nearly 300 employees for Talkie and Xingye, while the team size of MiniMax's productivity tool "Conch AI" does not exceed 100 people.

Almost every Talkie employee can list some examples of the fine design of Talkie to "Intelligent Emergence":

The UI specifically chose gold and black to be more luxurious; the star element should appear appropriately in various places, such as the eyes of some characters and various payment nodes; the dialog box with the character will automatically set parentheses, because users like to add "()" in the conversation to express their off-screen voice...

The chat interface of Talkie AI characters.

Iterating the product closely following user feedback makes it difficult for other products on the market to replicate the success of Talkie even if they copy its design completely.

"One thing I admire about Talkie is that it is very focused, and the functions are only to serve its target customers. Once the user group changes, Talkie's strategy will fail." An AI product manager of an AI unicorn evaluated, "Now many ToC AI applications are rather greedy, wanting to do all functions at once, resulting in low user activity and stickiness."

Difficult-to-Replicate Success

However, when an AI unicorn with a valuation of over 20 billion yuan has almost only Talkie as a "cash cow" product, there is also a hidden risk of total loss behind it.

The alarm bell rang for the first time in March 2023. At that time, MiniMax encountered a big hurdle: Glow, which had been online for only half a year, was removed from the app store.

Even though the official did not disclose the specific reason, and the artificial intelligence filing regulations had not yet been introduced at that time, many industry insiders and MiniMax employees believe that this is related to the chaotic user ecosystem of Glow at that time.

"80% of people will engage in borderline and pornographic content with the AI agent they create in Glow." A MiniMax product manager told "Intelligent Emergence".

Excessively catering to users has allowed MiniMax to taste the sweetness of the success of C-end products, but it has also planted hidden dangers for their development in the domestic market.

"Eating, sex, and human nature. Pornography is currently the most successful business model for AI." An investor once evaluated to "Intelligent Emergence". In contrast, AI products with significant overseas growth, such as Replika and Civitai, which are listed in the "Top 10 AI Applications with the Fastest Growth Rate in 2023 (Data Source: Writerbuddy)", are more or less tolerant of pornographic and borderline content.

To continue the sweetness, Talkie can only go to overseas markets with more relaxed regulations. However, MiniMax's domestic products have never been able to replicate the brilliance of Talkie.

For example, the domestic version "Xingye" that went online in September 2023 basically replicates all the gameplay of Talkie, and even the UI is more exquisite - initially, in order to adapt to the European and American markets with the highest payment capacity, the design of Talkie is not delicate. A former Talkie employee told "Intelligent Emergence" that the performance of foreign mobile phones is generally lower than that of domestic ones. Once there are too many engineering details, the application will lag, affecting the experience of European and American users.

But Xingye's popularity in China has not met MiniMax's internal expectations.

"In order to pass the filing, Xingye has spent a lot of effort setting up a firewall for sensitive words." The aforementioned MiniMax product manager mentioned, "But this greatly affects the product experience. Sometimes users are having a hard time in life and want to curse, but the AI advises you to be positive."

From the data results, female users have a higher willingness to pay for companion products than male users. A former Talkie employee mentioned to "Intelligent Emergence" that the male-female user ratio of Character.AI is 2:8, that of Talkie is 4:6, but that of Xingye is 6:4, "There is a gap in user payment willingness."

At the same time, Talkie overseas is not something that can rely on the market once and for all.

Anxiety is the emotion that many MiniMax employees can feel from Yan Junjie since 2024.

"The risk of a single source of revenue is too high. Currently, Talkie can be said to be the heart of MiniMax." An employee of Conch AI told "Intelligent Emergence". She mentioned that the US ban on TikTok has greatly shaken the senior management, fearing that Talkie will become too popular in the US and suffer the same fate.

This concern is not without reason.

On October 17, 2024, a Bloomberg report pointed out that as Chinese AI applications become more popular in the US, they are more likely to face the same geopolitical issues as TikTok.

This year, MiniMax has spent a lot of energy rectifying the content of Talkie. Several MiniMax employees mentioned to "Intelligent Emergence" that through some engineering means, Talkie's response to borderline conversations has gradually become less active.

"In the long run, borderline products cannot retain users. These users often use the application only for borderline content, and it's the same on another platform, with low stickiness." A Talkie product manager mentioned to "Intelligent Emergence".

Replicating another Talkie internally has also been put on the agenda of MiniMax.

The aforementioned Conch AI employee mentioned to "Intelligent Emergence" that in the middle of 2024, the company once required Conch AI, as a productivity tool, to add the companion attribute of independently creating AI characters - even though this decision has not produced the expected effect for the time being.

Several informed sources disclosed that the positioning swing between the product as a production tool and a companion tool has led to the collective departure of the product operation team of Conch AI; with strong competitors such as Kimi and Doubao, the user retention of Conch AI is not ideal at present.

At the same time, MiniMax has also put its eggs in different baskets and started to explore new markets outside Europe and the United States for Talkie.

An informed source disclosed to "Intelligent Emergence" that in March and April 2024, Yan Junjie mentioned at the weekly all-staff meeting that Talkie should start to layout the Southeast Asian market, and the product design and functions need to be adapted to the market.

"The growth of Talkie in the English-speaking market has entered a stable period, and there are still many opportunities in the small language markets in Asia." He mentioned that the advantage of Southeast Asia also lies in lower channel and marketing costs, and it is also more tolerant of nurturing products.

Continuing the growth story in the AI era is the key for MiniMax to secure a place in the highly competitive AI track.

Welcome to communicate!