Chinese startup DeepSeek is intensifying its work on a new generation of artificial intelligence designed to compete directly with American giants such as OpenAI, Anthropic and Microsoft. The Hangzhou-based company plans to launch advanced agent software by the end of 2025. The project aims to allow users to delegate complex, multi-step tasks to the system, which will gradually become more autonomous by learning from its previous actions. This technology aligns with the global trend of developing so-called AI agents — systems considered the next stage of AI evolution, capable not only of answering questions but also of independently carrying out sophisticated operations in both digital and real environments. DeepSeek’s initiative is seen as a response to the rapid progress of rival solutions in the US and China, where companies are racing to release next-generation platforms. According to sources, the company’s founder Liang Wenfeng is pressing his team to complete the software in the fourth quarter of this year despite market pressure and high investor expectations. The company’s ambition is to replicate the success of its R1 model, which in January surprised the global tech industry by matching or even outperforming some OpenAI products in tests, despite development costs estimated at only a few million dollars, reported The WP Times citing Bloomberg.
According to insiders, the startup is working on an AI model capable of handling multi-step actions on behalf of the user with minimal guidance. The system is also designed to learn and improve based on previous interactions. Founder Liang Wenfeng has set the goal of unveiling the product in the final quarter of this year.
DeepSeek joins a broader industry shift toward creating “AI agents,” regarded as the next stage of artificial intelligence. The tech sector is awaiting a successor to the R1 model, whose release in January triggered a global reaction. That platform, built for just several million dollars, managed to equal or surpass OpenAI’s tools in benchmarks.
Since R1, DeepSeek has limited itself to minor updates while US and Chinese competitors released new models at a rapid pace. Local media attributed delays in the new version to Liang’s insistence on perfection as well as possible technical issues. At the same time, he continues to run his profitable investment firm High-Flyer Asset Management.
Unlike Alibaba Group Holding Ltd. or Tencent Holdings Ltd., which frequently roll out new AI models, DeepSeek has taken a more cautious approach. Alibaba’s Qwen models have gained broad popularity, while Western rivals — OpenAI, Anthropic and Microsoft Corp. — have already launched their own agent software. Another Chinese-rooted startup, Manus AI, has announced the development of a general-purpose AI agent.
The long-term goal for DeepSeek, as well as the wider industry, is to create highly autonomous systems capable of planning and executing complex tasks independently — from writing code to organizing travel. For now, however, such agents still require significant human supervision.
Read about the life of Westminster and Pimlico district, London and the world. 24/7 news with fresh and useful updates on culture, business, technology and city life: US court ruling shapes the future of the internet: Google scores major victory
