A year after shaking up the world of artificial intelligence (AI) by unveiling a conversational agent that is much more economical than its American competitors, Chinese company DeepSeek launched a new, more efficient AI model based on open source on Friday.
Published on: 04/24/2026 - 06:55 Modified on: 04/24/2026 - 07:57
This is the thing that prevents artificial intelligence (AI) from going in circles. The company DeepSeek, a jewel of Chinese tech, announced on Friday April 24 the highly anticipated launch of a new AI model, which it presents as economical and one of the most efficient in the world.
In January 2025, the start-up from Hangzhou, a city in eastern China, stunned the world and reshuffled the AI cards with the release of a conversational agent competing, at a lower cost according to it, with its American competitors Gemini, ChatGPT or Claude.
The global tech industry has been watching for several weeks for Friday's announcement, a barometer of Chinese ambitions in the sector.
“Today, the preview of our newest model series, DeepSeek-V4, is officially live and released as open source,” the company said in a statement on social network WeChat.
This new model is offered in two versions: DeepSeek-V4-Pro and DeepSeek-V4-Flash, the latter being presented as more economical.
“Compared to the previous generation, the agent capabilities of DeepSeek-V4-Pro have been significantly strengthened,” the company said.
“Among the best in the world”
One of the main new features is the processing capabilities.
"DeepSeek-V4 has an ultra-long context of 1 million characters," the company said, meaning this new model can memorize and understand a million characters at once.
Exemple concret : en lui donnant des centaines de pages de texte, l'équivalent de plusieurs romans, il serait capable de répondre à des questions précises sur l'ensemble sans rien oublier entre le début et la fin.
Also read: DeepSeek, the “made in China” threat to American AI giants
This capacity is usually very computationally intensive, but DeepSeek claims to have made the process faster and cheaper.
The long context capabilities of the new model are thus “among the best in the world,” DeepSeek said on Friday.
“Welcome to the era of low-cost, million-character context,” she said on social network X.
In January 2025, DeepSeek's R1 conversational agent, endowed with extraordinary reasoning capacity, shook Wall Street, causing a fall in American technology stocks.
Sino-American rivalry
Presented as multimodal, that is to say capable of generating texts, images and videos, DeepSeek-V4 has the potential to once again shake up the titles of its American rivals, sector experts told Agence France-Presse (AFP).
The American company OpenAI, owner of ChatGPT, unveiled its own new AI model on Thursday, which claims to be the most advanced on the market.
GPT-5.5 is the next generation of the model on which ChatGPT is built, a generative AI interface now used by nearly a billion people.
Also read: When a disaster scenario on AI shakes Wall Street
The DeepSeek-V4 model can work with the latest chips made by Chinese giant Huawei, the startup said in a technical report.
Such a development marks a decisive step for China, which still does not have high-end chips from Nvidia, despite an easing of export restrictions by the United States.
With AFP
Keywords associated with the article
The content you are trying to access does not exist or is no longer available.
