DeepSeek, a Chinese start-up founded in May 2023 in Hangzhou, has quickly established itself as a major player in the field of artificial intelligence (AI), specifically in massive language models (LLM). A subsidiary of the hedge fund High-Flyer, DeepSeek is led by Liang Wenfeng and aims to compete with American AI giants by offering innovative and competitive open-source solutions.
DeepSeek specializes in developing massive language models capable of performing complex tasks through advanced reasoning capabilities. Since its launch, the company has introduced several notable models, including DeepSeek-V3, a model with 671 billion parameters that has been pre-trained on a vast dataset and stands out for its performance and drastically reduced training costs. This model competes with top American models, such as GPT-4o or Claude 3.5 Sonnet, despite limited hardware resources.
In January 2025, DeepSeek made waves with the launch of DeepSeek-R1, a first-generation reasoning model that disrupted the tech ecosystem due to its impressive performance and reduced training costs. This model was quickly adopted by the Chinese automotive industry for applications in driver assistance and enhanced interaction between drivers and vehicles.
DeepSeek continues to challenge tech giants with regular updates to its models. In May 2025, the company launched an update of its DeepSeek-R1 model, named DeepSeek-R1-0528, thereby enhancing its reasoning, logic, mathematics, and programming capabilities. This update allows DeepSeek to approach the performance of flagship models from OpenAI and Google while increasing the reliability of its responses through a significant reduction in hallucination rates.
Simultaneously, DeepSeek has begun distilling its models into lighter versions to make its solutions accessible to a broader audience, particularly developers with limited hardware resources. This strategy aims to democratize access to advanced reasoning capabilities without requiring expensive infrastructure.
DeepSeek has established itself as a serious alternative to American proprietary solutions, notably through its open-source approach that fosters collaborative innovation. By releasing its models under the MIT license, the company allows the research and developer community to freely access its technologies, thus stimulating innovation and the evolution of the open-source AI ecosystem.
The start-up also benefits from the support of the Chinese government, which sees it as a key vector for achieving technological self-sufficiency in the face of U.S. restrictions on the export of strategic components. DeepSeek is part of China's national strategy to become the global leader in AI by 2030.
DeepSeek has recently been in the spotlight with the temporary suspension of its chatbot in South Korea due to data privacy concerns. Although this highlighted some regulatory challenges, it has not dampened the enthusiasm around its technologies, especially in China where the DeepSeek-R1 model has been widely adopted in key sectors such as justice, cybersecurity, and public administration.
As rumors intensify around the imminent launch of DeepSeek-R2, the company seems well-positioned to continue challenging American giants and play a central role in the evolution of AI on a global scale. This upcoming model is expected to offer extended multilingual support and multimodal capabilities, paving the way for new applications in content creation and data analysis.
In conclusion, DeepSeek stands out for its ability to innovate rapidly and offer competitive solutions in a market dominated by tech giants, thereby strengthening China's position in the global race for artificial intelligence.
12 articles liés à cet acteur
At the end of January, Mistral AI introduced Small 3, announcing upcoming models with enhanced reasoning. Now, Mistral unveils Magistral, its first la...
The Chinese start-up DeepSeek has updated its R1 model, improving its performance in reasoning, logic, mathematics, and programming. This update, whic...
Meta AI is the most intrusive in collecting personal data, surpassing Google Gemini, according to a study by Surfshark. Meta AI collects 32 types of d...
OVHcloud announces the official launch of AI Endpoints, a new serverless cloud solution designed to facilitate the integration of artificial intellige...
As U.S. restrictions on the export of strategic components tighten, China is doubling down on its efforts to assert technological autonomy in artifici...
Launched last January, DeepSeek R1 quickly shook up Silicon Valley and the AI ecosystem, including Nvidia, due to its performance and lower cost. The...
On April 5, Meta unveiled the first two versions of Llama 4: Scout and Maverick. These open models, designed to be natively multimodal, can process te...
AGI is seen as the 'holy grail' by companies like OpenAI or DeepSeek, offering both opportunities and risks. Google DeepMind proposes a collaborative...
The Chinese start-up DeepSeek has quietly launched DeepSeek-V3-0324, an update of its eponymous open-source model DeepSeek-V3. This new version, with...
The Canadian unicorn Cohere recently unveiled "Command A," the latest version of its flagship model. Specifically designed to meet enterprise needs, t...
With the launch of R1, DeepSeek not only created a shockwave in Silicon Valley but also intensified competition within the Middle Kingdom.
After the spotlight on DeepSeek, China creates a new wave in the world of artificial intelligence with Manus AI, an autonomous agent that disrupts the...
