Open source + free is a "double-edged sword."

Author: Wang Lu

Image source: Generated by Wujie AI

AI seems to have become the "lifeline" for major companies.

Whether it's the highlight data in financial reports or the sporadic positive news, it all revolves around AI.

For example, in Baidu's mixed financial report for 2024, the highlights are basically provided by AI:

The daily call volume of the Wenxin large model continues to grow rapidly, increasing 33 times in a year to 1.65 billion. Baidu Wenku has over 40 million paid users, ranking second globally and first in China.

Alibaba has also made a three-pronged attack at the beginning of the year thanks to AI:

First, influenced by DeepSeek, Alibaba's open-source large model Tongyi Qianwen (Qwen) gained attention; then the latest model Qwen2.5-Max was rated as outperforming DeepSeek V3; subsequently, it announced a partnership with Apple for AI business, leading to a surge in stock prices.

However, since DeepSeek broke out nearly 40 days ago, the anxiety faced by major companies in AI has outweighed the gains. After all, each company has invested a significant amount of human, material, and financial resources, yet the standout product has come from a startup team. Recently, DeepSeek also publicly revealed explosive news—its cost-profit margin is as high as 545% (theoretical profit), with profits theoretically reaching 3.46 million yuan per day.

In light of various impacts, major companies are changing their strategies, joining forces instead of competing, announcing their integration with DeepSeek, and shifting their large models from closed-source to open-source, even going so far as to cut off a commercialization path by offering C-end products for free.

But can this wave of actions truly cure the AI anxiety of major companies?

How are major companies performing in AI?

Before the emergence of DeepSeek, major companies' approach to AI was characterized by high investment and a focus on leveraging their own advantages to create products.

Large models are seen as the infrastructure of the AI industry, with internet giants (Baidu, Tencent, Alibaba, ByteDance, Kuaishou, etc.), consumer electronics manufacturers (represented by Huawei), and intelligent voice manufacturers (like iFlytek) all launching their self-developed large models. Compared to startup companies like the "AI Six Little Tigers," major companies have the advantage of stronger financial and talent reserves.

From the overall technological iteration speed in the AI industry and the publicly available information from various companies, there is no fundamental difference in the underlying technology of major companies' large models, but there are differences in entry timing, model positioning, and market strategies, as detailed below:

These three differences represent, to some extent, the early attitudes and positioning of major companies towards AI.

For instance, the timing of large model releases indicates that an early release signifies that the company has made earlier investments and technological accumulations in the relevant field and can respond quickly, but the risk is that the technology may not be fully mature, leading to relatively higher costs in R&D and market promotion.

From the table, Huawei was the earliest, but it should be noted that although its foundation is also based on the Transformer architecture, it is fundamentally different from ChatGPT-style dialogue, belonging to the "industry-specific" direction of AI large models (ChatGPT-style is general intelligence). If focusing on general intelligence large models, Baidu acted the earliest, launching the Wenxin Yiyan large model invitation test in March 2023 (not fully open).

However, the timing of the launch is not the core factor in measuring the quality of a model.

The business layout of major companies determines the application direction of large models, leading to different model positioning, which technically stems from the training data of each company.

Baidu Wenxin Yiyan mainly relies on internet text data; Alibaba Tongyi Qianwen uses multimodal data including text, images, and audio; Tencent's Hunyuan relies on social network and user behavior data; ByteDance's Doubao derives about 50%-60% of its data from its own businesses (Douyin, Toutiao); Huawei's Pangu large model utilizes various data including industrial, meteorological, text, and image data.

This also results in different advantageous scenarios for each large model. For example, Wenxin Yiyan excels in long text processing and multilingual dialogue; Hunyuan is superior in social scenarios; Doubao leads in content generation and precise recommendations; Tongyi Qianwen responds faster in e-commerce recommendation scenarios; Pangu has excellent execution speed and generalization ability, efficiently handling large-scale tasks.

It is not difficult to see that the advantageous fields of each large model reflect the core business of each company.

Finally, looking at market strategies, which to some extent reflect major companies' judgments on their own capabilities and industry trends, the observable content can be roughly divided into two parts: open/closed source and whether TO C products are free.

ByteDance, Kuaishou, iFlytek, and Huawei are still adhering to closed-source, while Baidu, Tencent, and Alibaba have chosen to open source most of their models. In TO C applications, Baidu, Tencent, and Alibaba have opted for a free route, while ByteDance, Kuaishou, and iFlytek mostly provide limited free quotas.

Alibaba has already reaped the benefits of open source; the latest open-source large model ranking released by the open-source AI platform Hugging Face shows that all top ten open-source large models are derived from Alibaba's Tongyi Qianwen.

In TO C products, Doubao, which insists on being free, has seen the most significant growth over the year. According to the AI product rankings, in January 2025, Doubao ranked first in the domestic club of tens of millions of monthly active users, with 78.61 million, far surpassing other major company applications.

However, what everyone is more curious about is the overall capability ranking of major companies' large models. According to several industry analysts, currently, the top large models from major companies are primarily closed-source, making it challenging to assess each company's capabilities in the absence of fully transparent information.

Frost & Sullivan's report "2024 China Large Model Capability Assessment" points out that Baidu Wenxin Yiyan, Tencent Hunyuan, Alibaba Tongyi Qianwen, and other major company large models are all in the first tier, indicating that they have relatively comprehensive technical capabilities and a larger user base. However, it does not provide a clear judgment on which company has superior overall capabilities.
Software engineer Qin Xiang stated that there are differences among companies in terms of technical architecture and training data. For example, from a technical architecture perspective, model scale and parameter count are important indicators of the complexity and capability of large models. Generally, the larger the scale and the more parameters, the stronger the model's learning and expressive capabilities. For instance, DeepSeek-R1 is referred to as a parameter giant, boasting 671 billion parameters that create a vast knowledge repository.

He noted that from this dimension, among the large models in major companies, those with deep reasoning capabilities, such as Wenxin Yiyan, are stronger. However, when looking at vertical domain capabilities, Wenxin Yiyan cannot match Tongyi Qianwen, as the latter has developed and launched eight vertical domain models based on its own technology.

In summary, the advantages of each large model are different, making it difficult for any one company to dominate all dimensions over others.

Four Major Changes in Major Companies After DeepSeek's Breakout

The emergence of DeepSeek has prompted major companies to reassess their AI strategic layouts. Based on the latest developments from various companies and industry insiders, there are four major changes.

First, from closed-source to open-source, which is the most significant change.

More than one industry insider pointed out that the popularity of DeepSeek is closely related to open source.

Discussions about the open and closed source of large models have been ongoing both domestically and internationally. Baidu's chairman, Li Yanhong, was once a staunch supporter of closed source, believing that closed source is superior to open source in maintaining technological leadership and business models.

Qin Xiang analyzed from a technical perspective that open source means that core code is public, allowing competitors to quickly replicate the technological path. Major companies initially chose closed source mainly to protect intellectual property and commercial barriers (like OpenAI's early decision not to open source GPT-3).

However, he found that under the influence of DeepSeek, major companies have changed direction, leaning towards achieving long-term benefits through ecological binding (e.g., Tencent Hunyuan open-sourcing video models to attract developers to use its cloud services) rather than solely relying on technological secrecy as before.

Now, Baidu has announced that the Wenxin large model 4.5 series will be fully open-sourced by the end of June 2025. As of now, most models from Baidu, Alibaba, and Tencent have either been open-sourced or announced plans to do so.

Second, the business focus has shifted from TO B to "dual-line parallel."

Qin Xiang explained that there are three main monetization methods for large models: value-added services, data monetization, and compliance services, with value-added services accounting for the largest share, relying on enterprise-level customization and API call revenue. He revealed that the annual fee for Baidu Wenxin Yiyan's enterprise version exceeds 10 million yuan, while Alibaba Cloud's Tongyi Qianwen provides customized customer service systems for government and enterprise clients, with single project contracts reaching hundreds of millions.

This means that major companies currently still rely primarily on B-end profits, but many have recently begun to emphasize the promotion of TO C applications, shifting to a "dual-line parallel" approach of TO B and TO C.

Image source / Pexels

For example, Tencent has increased its promotion of Yuanbao, integrating it into the WeChat nine-grid interface, gaining a strong traffic entry, while also advertising through multiple channels, including significant placements on Douyin, Bilibili, and Zhihu, in addition to promoting within Tencent's ecosystem.

According to App Growing data, major AI products from major companies are listed among the top 20 AI tools by advertising intensity in February (Huawei is not included as it does not have TO C products). The largest spender was Tencent's Yuanbao, which accounted for 46% of the total advertising expenditure in February, nearly matching the total from the past nine months, surpassing ByteDance's Doubao.

Additionally, Alibaba is also recruiting extensively for TO C business-related talent.

Industry insiders believe that the pressure from DeepSeek's open-source + low-cost APIs has led major companies to seek more commercialization opportunities in TO C.

The third change is that TO C applications have shifted from paid to free.

DeepSeek is user-friendly and free, and following its popularity, both Baidu's Wenxin Yiyan and OpenAI's upcoming GPT-5 have announced they will be free for users.

"The goal is to attract more users and increase market share," Qin Xiang stated. More user feedback can further optimize model performance, thereby enhancing B-end service capabilities and allowing for higher fees for customized enterprise models.

The fourth change is from heavy investment to cost reduction and price wars.

In the past few years of the "hundred model war," domestic and international AI large model companies have spent tens of billions or even hundreds of billions of dollars, while DeepSeek trained its DeepSeekR1 model, comparable to OpenAI's capabilities, with just $5.576 million in GPU costs, prompting major companies to reflect.

More than one industry insider has indicated that major companies began cost reduction in the second half of last year, but the emergence of DeepSeek has accelerated this trend.

Qin Xiang has clearly felt that since last year, competition among large models has shifted from "technology first" to "cost + ecosystem." For instance, the API price for Doubao 1.5 Pro significantly decreased after its release in January last year, and in December, ByteDance reduced the price of its visual model by 85%, pushing the industry into the "penny era."

In February of this year, two veteran Baidu employees engaged in a "remote battle" over large model pricing. Shen Dou, president of Baidu's Intelligent Cloud Business Group, pointed out at a meeting of the Baidu Intelligent Cloud Business Group (ACG) that there is a "malicious price war" in the domestic large model industry, specifically naming Doubao. Subsequently, Tan Dai, president of ByteDance's Volcano Engine, responded on social media, stating that price reductions are an inevitable result of technological progress.

DeepSeek has also been active, having just announced the end of its API discount period, and on February 26, it announced a "limited-time price reduction," with DeepSeek-V3 dropping to 50% of its original price and DeepSeek-R1 going as low as 25%, with reductions of up to 75%.

The pressure on major companies has increased.

Can Major Companies Win Back Their Home Turf with Free and Open Source?

According to industry insiders, among the four major changes, the most significant impact on major companies currently comes from open source and free offerings.

First, let's look at open source.

Liu Cong, an expert in the field of large models, pointed out that before DeepSeek emerged, both foreign companies like OpenAI and domestic giants either chose to keep everything closed source or only open-sourced some models (not the best versions). DeepSeek, however, chose to open source its most powerful reasoning model, DeepSeek-R1, which has excited industry professionals.

However, open source also faces some revenue losses and technical risks.

Dr. Wei Liang, an AI PhD, stated that open/closed source represents two different business models and development approaches: indirect and direct monetization. A typical open-source representative among domestic giants is Alibaba's Tongyi Qianwen large model, which promotes commercial cooperation by adapting to manufacturers, a choice based on its own ecosystem.

However, many major companies initially positioned their large models as technology-driven, viewing them as productivity tools. For example, OpenAI, Baidu, Huawei, and iFlytek consider subscription fees for large models to be a significant source of revenue, and choosing to open source will undoubtedly impact their earnings.

Open source also faces risks of malicious attacks and community maintenance issues. For instance, with code being public, malicious attackers can analyze the code to find vulnerabilities, potentially attacking systems that use these models.

Subsequent community maintenance is also a concern. Qin Xiang noted that open source requires continuous resource investment to maintain the developer community (such as providing documentation, technical support, and version updates); otherwise, it may lead to a fragmented technical ecosystem. He explained that if developers modify the code themselves and create multiple branches (like Linux branches Ubuntu and CentOS), it would increase the difficulty of unifying technical standards, leading to "technical fragmentation."

Image source / Pexels

Some industry insiders bluntly stated that even if major companies open source, their attractiveness is limited.

The purpose of open source is to attract technical developers and partner companies, encouraging them to use their large models for technical iteration and application development. However, Dr. Wei Liang believes that "currently, the open-source efforts of various companies seem to have advertising motives."

"What can be seen in open source are the reasoning methods and parameter weights of large models, but more importantly, the data filtering methods and model training techniques are not disclosed by any company, making it difficult for ordinary developers to perform technical iterations," he stated.

It is worth noting that open source does not equate to being completely free; users must still comply with the open-source agreements provided by large model vendors, which include "payment clauses."

For example, Dr. Wei Liang uses Alibaba's Tongyi Qianwen large model for some AI applications. After running the technology through Qianwen, if he wants to further customize and adapt it for enterprises, he needs to contact the staff. He also revealed that the open-source agreement may have restrictions based on company size, such as requiring payment once the number of employees reaches a certain threshold.

Now, let's look at the impact of free offerings.

The purpose of major companies adopting a free strategy is to quickly capture the C-end market. A prominent example is Doubao, which has always been free for users. According to QuestMobile data, as of February 9, 2025, Doubao's average daily active users (calculated as the average number of active users per day from February 3 to February 9) reached 18.45 million, second only to DeepSeek and higher than Kimi, Wen Xiaoyan, Tongyi, and Yuanbao.

However, the significance of being free is still uncertain among industry professionals. This is partly because users have low loyalty to tools like chatbots and also because domestic users do not have a strong willingness to pay.

"Even for AI-generated video tools that require payment, most domestic applications attract users by offering free credits," stated one industry insider, who believes that Doubao's success among similar general-purpose AI products is not only due to being free but also closely tied to ByteDance's strong market promotion.

Qin Xiang believes that DeepSeek's catfish effect is forcing major companies to shift from a technology competition to a comprehensive contest of cost and ecosystem. The open-source and free strategies are a "double-edged sword" in responding to competition and building ecosystems, and even though these measures may reduce their own earnings in the short term, they have no choice but to adopt them.

The catfish effect triggered by DeepSeek is far from over.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。