GuShiio.com Article source: New Entropy
Image source: Generated by AI
The chain reaction about DeepSeek continues.
As the direct beneficiaries of this round of DeepSeek’s outbreak, major manufacturers have recently taken out real money and continued to increase their AI.
On February 24, Alibaba CEO Wu Yongming announced that in the next three years, Alibaba will invest more than 380 billion yuan in building cloud and AI hardware infrastructure, totaling more than the total of the past ten years. Tencent even announced the reorganization of its business group, connected to DeepSeek, and began to be all-in-AI.
The popularity of the big model brought by DeepSeek is like a “disaster” for other players in the industry. Data decline, business shrinkage, and even strategic reorganization have become the norm. The protagonist injured in this “disaster” is undoubtedly Kimi.
According to industry analysts, domestic large models are used on the c-terminal, experiencing a state of decline and decline. The fire at DeepSeek has also caused Kimi and bean bag users to fall back from their peak. A typical representative is Tencent Yuanbao, with the help of its parent company, quickly occupied the top spot in Apple’s downloads and surpassed Kimi in daily life. It seems that it is not difficult.
For Kimi, what’s even more difficult isHow to turn users ‘minds.Even though the gap between k1.5 and DeepSeek may not be large, users ‘perception is that DeepSeek is not only traffic, but also ecologically empowered. Whether it is search engines, mobile phones, or cars, they have all been connected to DeepSeek, and a large model ecosystem similar to “Android” is gradually taking shape. For Kimi, the road ahead is undoubtedly destined to be bumpy.
Famous investor Zhu Xiaohu once mentioned in an interview that after the explosion of DeepSeek, even big factories such as Doubao were difficult to catch up. Once the user’s mind is formed, it will be too difficult to conduct market education.
Can Kimi make a comeback in the future? Judging from the company’s development path, it is not impossible. As the company that once followed OpenAI technology the fastest, Kimi made several models that followed OpenAI technology, such as Kimi Discovery Edition and k0-math last year. Yang Zhilin himself also said in an interview that the future of the large model lies not only in strengthening learning, but also in multimodal capabilities. This seems to be similar to OpenAI.
This means that if OpenAI launches disruptive innovation again, Kimi will still be the fastest-following existence possible in the country. But whether this can be used to change users ‘minds requires a question mark.
From a product and ecological perspective, Kimi is undoubtedly facing many difficulties at the moment. Internally, the battle between founders and investors seems to be continuing. Externally, in the face of the continuous acceleration of chasers, what application solutions can Kimi come up with? Questions such as launch, open source, and even multimodal capabilities are all coming to Kimi. How to answer them is testing the foundation of this company.
Kimi didn’t have a good New Year
When metaphysics, stocks, and countless hot searches are all beginning to be tied to DeepSeek, AI applications are entering an unprecedented explosion period.
The current leader in the right to speak undoubtedly belongs to DeepSeek. Whether it is Liang Wenfeng’s every move, every paper published by DeepSeek, or corporate cooperation, and industry popularization, they can easily attract attention.
The audience of AI applications is growing rapidly, but as a former track star, Kimi obviously has difficulty receiving this dividend.
On the contrary, during the Spring Festival, Kimi’s internal and external problems are gradually being amplified.
Internally, the long-shelved arbitration case did not complete the settlement as expected, but entered the next process of the arbitration case. According to covert reports, the two parties to the Kimi arbitration case, the old shareholders of Circular Intelligence and Yang Zhilin, have completed payment at the HKIAC (Hong Kong International Arbitration Center) at the end of January and late February respectively, and the court formation has also been completed. Zhang Yutong, the more critical protagonist behind the entire incident, may be sued separately.
According to another investor close to the negotiations, after investor Zhu Xiaohu publicly denounced and Yang Zhilin responded to an open letter, Yang Zhilin once came out to negotiate with the old shareholders and expressed his willingness to sell some shares, and a small portion of them also came from Zhang Yutong’s transfer. The implication is that Kimi is not willing to cut off with Zhang Yutong.
Obviously, the arbitration between Kimi’s founding team and Circular Intelligence will continue in the short term. Whether the arbitration will affect Kimi’s financing and management will become the focus of external concerns.
While internally subject to arbitration, Kimi’s products also seem to be under pressure at different levels. It has become an indisputable fact that the data has dropped significantly. During the Spring Festival, it was rumored online that Zhang Yutong’s circle of friends published relevant content about the further growth of Kimi users, but from the data disclosed by other institutions, it seemed to be inconsistent. According to the AI application list released by Aicpb.com, Kimi has dropped to third place in January, and its monthly active number has also dropped from 36 million announced by Yang Zhilin in October last year to 19.43 million.
But for the outside world, the most worrying thing is still the large-scale shift of users ‘minds. To a certain extent, Kimi has become DeepSeek’s “equivalent” product. Although from the product point of view, k1.5 has remarkable innovation in the direction of long text and multimodal, due to its “untimely timing”, it has gradually been drowned in DeepSeek’s voice, and it can be said that there is no splash. The same case also appeared in the recent “paper crash” incident.
On February 18, DeepSeek and Dark Side of the Moon published two papers on the same day. The former was submitted by Liang Wenfeng personally on February 16, and the latter, Yang Zhilin’s name rarely appeared in a key position as the author of the paper.
Reports show that unlike DeepSeek, which only published one paper, Dark Side of the Moon also released related code. Moreover, these codes have been verified by actual deployment for one year, and their effectiveness and robustness are guaranteed.
However, as far as external evaluations are concerned, discussions on Liang’s technical route are obviously more popular than Yang Zhilin’s MoBA. The reason is that on the one hand, Liang Wenfeng personally went into battle, on the other hand, he continued the shock of R1, and on the other hand, perhaps people are more eager to see the “rebels” and recognize his firm attitude.
Internally under arbitration pressure, external competition suppresses the decline in user numbers, and changes in user minds will further torture Kimi’s previous operating strategies. Whether to continue to invest in large amounts and what to support the next round of financing. Technically, when the accumulation in the “multimodal” direction of Yang Zhilin’s investment will be released will become a question that Kimi has to continue to think about.
Dachang follows the old path of Kimi
The narrative that DeepSeek exposes lies not only in low-cost training solutions, but also in people’s aversion to “pseudo-innovation.”
In the launch war that began last year, Kimi was undoubtedly the one who was most injured. Based on public information, Kimi’s monthly activity in a report released by QuestMobile in September was 6.25 million, and by November, Yang Zhilin claimed that his monthly activity reached 36 million.
According to appGrowning platform data, starting from March 2024, Kimi has spent tens of millions of yuan in advertising almost every month. The advertising amount in October even reached 220 million yuan. At that time, Kimi was also on a hot search for “burning 100 million yuan in 20 days.” In November, Kimi’s investment amount once again reached 200 million yuan. Based on a simple addition of the platform information, Kimi, with a total of 36 million users, spends nearly 500 million on advertising. (Only added from media information, for reference only)
But by January 2025, Kimi’s monthly activity has dropped to approximately 19 million. Obviously, under DeepSeek’s super AI application narrative, it is difficult to stop Kimi user loss.
Of course, this part of the loss of users may be related to Kimi’s beginning to rethink investment and cut budgets. According to reports on the interface, Kimi recently decided to significantly reduce its product launch budget, including suspending the launch of multiple Android channels and cooperation with third-party advertising platforms.
But in any case, Kimi’s failure in operational strategy is having a ripple effect.Among them, the most obvious example comes from Tencent’s personal death.
ADX data shows that before February 5, the amount of material released by Kimi, Doubao and Tencent Yuanbao was significantly stratified, and there was a large gap. After that, the amount of material released by Kimi gradually declined, and became the lowest among the three around February 15. So far, it has maintained the daily amount of material no more than 1000 groups.
After February 5, the release intensity of bean buns first increased to a certain extent, and then dropped sharply, maintaining a similar release intensity to that in late January. In comparison, the changes in Yuanbao are more obvious.
In the past, Yuanbao’s investment efforts have never been high, and Tencent has also been considered by the outside world to have vague strategies and slow actions in the field of AI models.
However, on February 15, the release of Yuanbao increased slightly, and on February 18, it increased sharply. So far, the highest number of materials released in a single day has exceeded 11,000 sets, making it the first among the three.In contrast, in the six days (2.18-2.23) in which the release of ingot has increased significantly, a total of 55,000 sets of materials have been released. The month-on-month release has increased by 12,000 sets of materials compared with the previous six days (2.9-2.14), an increase of 345.1%.
On February 22, Tencent Yuanbao surpassed Doubao and rose to second place in the free list for Apple in mainland China. Because Tencent’s applications are rich, this means that whether it is application linkage or mutual empowerment, it will have better results. To a certain extent, Yuanbao has followed Kimi’s path, and it will even become possible to overtake Doubao in the short term.
Falling behind two positions at a time, and at the same time, the giant has followed its own old path, obviously Kimi not only needs reflection, but perhaps also needs a thorough change. DeepSeek’s case seems to have taught Yang Zhilin that he should return to technology as the foundation. So from a technical point of view, does Kimi have any hope of turning the tables?
The comeback depends on both OpenAI and DeepSeek
To see Kimi’s future clearly, perhaps a clearer path lies in observing DeepSeek and OpenAI. Among them, DeepSeek will directly affect Kimi’s product rhythm. After all, if there is a large gap between the time and ideas of DeepSeek’s product launch conference, it means that Kimi will further lose competitiveness. OpenAI, as a so-called closed-source and technologically leading representative, determines the upper limit of Kimi’s path development.
According to the news, reports show that DeepSeek may release r2 before May. According to Microsoft, the company is preparing for the latest model of OpenAI. It is expected that GPT-4.5 is expected to be available as soon as next week. Microsoft expects GPT-5 to be released in late May. According to the analysis of information equality among industry observers, what DeepSeek needs to do is to continue to open source while others continue to close sources. r2 should be benchmarked against the full version of GPT o3, while DeepSeek V4 should at least benchmark GPT-4.5. Based on the V4+RL model, it should be benchmarked against the future so-called “GPT-5”.
Judging from this benchmark and pace, Kimi’s success or failure may occur in the next three to four months. If DeepSeek and OpenAI have completed a new round of updates, further occupying users ‘minds, and at the same time, participants in the DeepSeek camp such as Yuan Bao continue to increase their investment efforts, then for Kimi, the only remaining value may be the acquisition.
In terms of technical route, Kimi’s lead over DeepSeek lies in its multimodal capabilities. According to industry analysts, DeepSeek will only add multimodal capabilities until at least the V4 version. In contrast, Kimi seems to have more accumulation across multimodal aspects.
In Caixin’s interview, unlike many large model companies in the industry who regard multimodal capabilities as the next step towards AGI (General Artificial Intelligence) after language models, Yang Zhilin believes that reinforcement learning is more important. “The next step in the big model’s capabilities is thinking (reinforcement learning) and interacting (multimodal),Multimodal is necessary, but thinking determines the upper limit of the ability of large models.” He also emphasized that several multimodal models of the dark side of the moon are currently under internal testing.
On December 16, 2024, Kimi released the k1 visual thinking model. According to comments from the evaluation agency, k1 ‘s visual ability supports complex scenes, including unclear photos, mixed shots of multiple questions, and interference from handwriting. Even purely handwritten questions, k1 can accurately recognize and give the reasoning process, far exceeding the response ability of traditional AI.
Subsequently, on January 15, Kimi released the multimodal picture understanding model API. The new multimodal picture understanding model moonshot-v1-vision-preview improved the multimodal capabilities of the moonshot-v1 model series.
Then k1.5 appeared, showing more multimodal capabilities. However, these are obviously difficult to open the gap with DeepSeek.
But what I have to admit is that Kimi has certain competitiveness in catching up and copying OpenAI in the direction of multimodal capabilities.
totalIn conclusion, Kimi’s window of opportunity seems to be shrinking from all dimensions. Among the influencing factors, we must not only refer to the progress of OpenAI, but also fully find out the differences between OpenAI and DeepSeek. At the same time, internal contradictions seem to be a variable in the future.
The emergence of DeepSeek has, to a certain extent, completely rewritten the ecological niche of large model manufacturers represented by Kimi. What is certain is that if another technical breakthrough cannot be achieved, it means that the knockout round may occur at any time. As for whether there will be a real buyer, everything seems to be in the fog.
