Beijing, (EFE) .- The Chinese technological firm Depseek presented an updated version of its R1 reasoning model, which the company ensures that it has registered a prominent performance, according to various evaluation criteria in areas such as mathematics, programming and general logic.
The new model, which is already available on its website and application, has been the subject of a new training process with expanded computational resources, the company said in a statement in the Wechat social network.
The company indicates that this improvement has significantly increased its reasoning and depth capabilities in thought processes.
Deepseek points out that the updated model, called Deepseek-R1-0528, shows a reduction of approximately 47% in the errors rate in rewriting, summary and reading comprehension tasks, a phenomenon that brings to the sector head and which is known as ‘hallucinations’.
In addition, the company maintains that its performance is approaching that of international reference models such as O3, Openai, and Gemini 2.5 Pro, developed by Google.
The LiveCodeBench site, a comparative index prepared by researchers from California universities in Berkeley, the Massachusetts Institute (MIT) and Cornell University, locates R1-0528 in positions close to models such as O4 Mini and O3 of Openai in the code generation, collects the local newspaper Global Times.
Lee: South Korea says Deepseek transferred user data and messages without consent
According to this indicator, the Depseek model is above Grok 3 Mini, of the XAI company, and Qwen 3, from Alibaba.
Recent language models such as Deepseek and others presented by Chinese technology companies such as Baidu, Tencent or Bytedance have caused a great media and economical stir in recent months for being able to rival in their capabilities with their US competitors at a lower price.
Most of these companies have opted for open source models, allowing third parties to modify their base structure and adapt it to specific needs, a strategy applauded by Chinese authorities.
However, some experts in the sector have been skeptical of the sudden irruption of so many similar services simultaneously in China and there are doubts about the real global expansion capacity of the Chinese ‘chatbots’ for the censorship exerted by the authorities.
China already regulated in 2023 the services of AI to respect “the fundamental socialist values”, so that these platforms are prohibited “to generate content that threaten national security, territorial unity and social stability.
Inspy, discover and share. Follow us and find what you are looking for on our Instagram!
