DeepSeek is preparing a new AI model to compete with OpenAI - Bloomberg

Chinese startup DeepSeek is developing a new artificial intelligence model with advanced AI agent functions, Bloomberg's sources claim. This will allow the company to compete with U.S. players like OpenAI at the new frontier of technological advancement, Bloomberg notes. The release of DeepSeek's R1 AI model earlier this year brought down the quotes of U.S. AI companies.

DeepSeek is working on a model that can perform complex multi-pass tasks on behalf of the user, relying on only minimal instructions, Bloomberg's interlocutors said. The system should be able to learn from its own actions and become more efficient over time, they added.

Thus, DeepSeek joins the global trend of creating "real" AI agents, which are increasingly being called the next stage in the evolution of artificial intelligence, the agency writes. Its sources say that the company's founder Liang Wenfeng aims to unveil a new product in the fourth quarter. The market is waiting for the company's successor to the R1 model - a neural network, the appearance of which shook the global technology industry in January, Bloomberg explains.

The R1 platform, which mimics the process of human thinking, was rumored to have been created for only a few million dollars, but managed to compete on equal terms, and by a number of metrics - and even outperform OpenAI solutions in benchmarks. In January, this led investors to question the feasibility of multi-billion dollar investments in AI and began selling off shares in many AI-related companies, including Nvidia.

The development of the new agency AI underscores Liang's desire to maintain its leadership in an ultra-competitive industry, though it remains to be seen whether it will be able to replicate the success of R1, Bloomberg writes.

Since the release of R1, the startup has only released minor updates, while competitors have been introducing new releases one after another. Chinese media attributed the delay of the R2 model to Liang's efforts to perfect it despite his busy running an investment company, High-Flyer Asset Management. Some media speculated about technical glitches during the training or development phases of the R2.

DeepSeek did not respond to Bloomberg's requests for comment.

DeepSeek's plans to release a new model with a focus on agency reflect a broader shift in the technology industry, Bloomberg explains. In recent months, OpenAI, Anthropic and Microsoft have unveiled their own versions of AI agents to simplify personal and work tasks. Unlike traditional chatbots limited to a couple of short lines, the new generation of such systems are designed to handle more complex tasks, from organizing vacations to writing and debugging software code. In a recent analyst report, Goldman Sachs called AI-assisted business productivity improvement the next big step in the development of generative AI.

The goal of DeepSeek, and the industry as a whole, is to create increasingly autonomous AI systems capable of initiating and realizing complex actions in the real world with minimal need for human intervention. In practice, however, such agents still require quite a bit of human oversight, Bloomberg notes.

This article was AI-translated and verified by a human editor