The world of artificial intelligence (AI) has been rocked by an explosive controversy. OpenAI, known for its revolutionary innovations, has leveled serious accusations against DeepSeek, a promising Chinese startup that has rapidly gained notoriety in the global AI landscape. The central allegation? Intellectual property theft. The accusation, which surfaced in the last week of January 2025, claims that DeepSeek improperly used OpenAI model outputs to enhance its own AI systems.
The technique in question, known as “distillation,” is at the heart of the dispute. OpenAI alleges that DeepSeek leveraged the results generated by its most advanced models to train smaller, more efficient models. This practice, while not inherently illegal, is explicitly prohibited in OpenAI’s terms of service, especially when used to develop competing models. Evidence presented by OpenAI includes alleged screenshots indicating DeepSeek responses that are strikingly similar to those of ChatGPT, OpenAI’s language model.
Microsoft, OpenAI’s partner and server infrastructure provider, has also weighed in, claiming to have detected suspicious data movement originating from China, involving the mass extraction of data from OpenAI’s API. The American tech company suspects these accounts are linked to DeepSeek. Interestingly, DeepSeek gained prominence for developing AI models with remarkably superior efficiency, achieving 10 times the performance of other systems, partly by avoiding the use of Nvidia’s CUDA platform. Instead, the Chinese company reportedly used direct parallel execution of Nvidia threads, a complex and uncommon approach.
Amidst this turmoil, a new Chinese player emerges: Alibaba, which launched the Quen 2.5 Max model. While not a reasoning model, Quen 2.5 Max stands out for surpassing models like DeepSeek, Claude, and GPT-4o in specific benchmarks. Additionally, another Chinese model, Kim 1.5, was also released, with claims that it surpasses OpenAI’s own GPT-4o in certain aspects. This scenario points to fierce competition between China and the United States in AI development, while Europe seems to be focusing on other technological innovations.
Despite the accusations and controversy, DeepSeek also launched the Jan series models, focused on diffusion-based image generation. While the quality doesn’t yet compare to models like Stable Diffusion or Midjourney, the initiative represents another open-source model available for commercial use. This move reinforces the growing trend and importance of open-source AI models, which offer alternatives and democratize access to the technology.
