Meta's open-source large model, Llama3, seems to have received a "cold shoulder" in the market, a phenomenon that undoubtedly intensifies the competition between open-source and closed-source large models. According to reports from The Information, the world's largest cloud computing platform, Amazon AWS, has shown lukewarm interest in Llama3, while corporate clients appear to favor Anthropic's closed-source model, Claude.
Insiders at Microsoft also revealed that Llama is not their top choice, instead recommending companies with engineering and data science teams. The challenges Meta is facing may prompt them to form a dedicated AI sales team to better meet corporate needs. This all shows the difficult journey of open-source large models in the process of commercialization.
Image source note: The image was generated by AI, provided by the image licensing service Midjourney
From a market perspective, the actual performance and commercial returns of open-source models have clearly not met the expectations of corporate clients. In the choice between open-source and closed-source, major model vendors have formed very different positions based on their own technical routes and business strategies. So, how should companies navigate this dispute when choosing large models? Xin Zhou, General Manager of Baidu Smart Cloud AI and Large Model Platform, provided an in-depth analysis in an interview with the media, discussing the underlying logic and business strategies of open-source and closed-source, and forecasting future market trends.
Xin Zhou pointed out that open-source large models and open-source software are entirely different concepts. Open-source models do not disclose important information such as training source code, pre-training, and fine-tuning data, which affects model performance, and therefore cannot rely on community developers to improve their performance like open-source software. Taking Llama as an example, each improvement of the model comes from Meta's own training, not from developer participation. This is why open-source models face many obstacles in technical iteration.
When discussing "which is more expensive, open-source or closed-source models," Xin Zhou said that although open-source models seem free on the surface, giving the illusion of low cost, the application of large models does not rely solely on technology but includes an overall solution of "technology + service." In practical applications, to achieve the same effect as closed-source models, companies need to invest a lot of manpower, funds, and time, which may result in a higher overall cost.
Open-source and closed-source models also have different applicable scenarios. Xin Zhou believes that open-source models are more suitable for academic research and are not suitable for large commercial projects that require external services. In major projects where investments can reach millions or even tens of millions, closed-source models are still the preferred choice for enterprises.
Xin Zhou further analyzed the roles and business models of various vendors in the current large model market, pointing out three main types. The first are cloud service providers, whose business model still involves providing computing power resources and reducing costs through scale. The second are companies that are both cloud service providers and model providers, who hope to drive cloud business through model calls. The third are start-up model vendors, who face significant challenges in the market price war.
In summary, open-source models have significant disadvantages in both technology and business models. Lacking sufficient resources and good business model support, many open-source models struggle to sustain development. Although open-source has value in promoting academic research, in commercial scenarios requiring high precision and efficiency, closed-source models remain the wiser choice.