A Turning Point in the AI Race: Open-Source Models Take the Lead

In recent months, the focus in the world of AI has been on new flagship models from giants like OpenAI and Anthropic. However, an unexpected trend is emerging among industry experts and developers: instead of closed and expensive models, open-source and flexible technologies are capturing the market. This signals a shift in the balance of power in the AI race, as reported by Techcrunch.com reports.

According to data from the Hugging Face platform, 41% of model downloads in the spring of this year were for Chinese open-source models, surpassing US models. On the OpenRouter platform, all six of the most popular models are open solutions provided by Chinese companies such as Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Even the popular Claude Opus 4.7 model has fallen to seventh place on this list.

Why are companies choosing open models?

Statistics from the Vercel platform show that open-source models are now handling the bulk of infrastructure workloads, while closed models are reserved only for high-cost, premium-level tasks. In June, nearly one-third of all AI requests were processed via open models. This trend demonstrates that businesses are striving to optimize costs.

Clem Delangue, CEO of Hugging Face, believes that in the coming years, the most powerful "frontier" models will be used only for experiments and highly complex, high-value tasks. Daily production processes and core workloads will shift to companies' own proprietary models or open-source solutions, challenging the concept of a "single, smartest model" created by large laboratories.

Companies do not want to entrust their core capabilities to another organization in the form of "black box API" systems, where internal mechanisms are unknown and uncontrollable. Owning one's model not only ensures security and control but also reduces the massive costs associated with scaling AI.

China's technological offensive

China's AI labs are releasing new, powerful, and open-source models every few months. For example, Beijing-based Z.ai recently released the GLM-5.2 model. This model can compete with Anthropic's latest developments in coding and identifying vulnerabilities in security systems. Most importantly, configuring and running such models is much cheaper than using US closed systems.

Currently, half of the Fortune 500 companies are using the Hugging Face platform to deploy their own proprietary and open-source models. A new repository is created on the platform every seven seconds, highlighting the high level of activity in the field. Today, there are nearly 3 million public models and over 1 million datasets on the platform.

In conclusion, the future of AI is increasingly likely to lie not in the hands of a few tech giants, but in an open ecosystem developed collaboratively by programmers worldwide. This opens doors for developing nations, including Uzbekistan, to create their own AI solutions.