DeepSeek demand shrinking: Baidu founder notes shift away from text-only AI

Chinese tech major Baidu launched Ernie 4.5 Turbo and X1 Turbo, which feature multimodal capabilities including text, image, audio and video

Baidu Co-founder Robin Li said the demand for text-only generative AI models — such as those developed by industry frontrunner DeepSeek — is “shrinking”, in a direct swipe at one of China’s most high-profile AI companies.

 

Speaking at Baidu’s developer conference on Friday, as reported by the Financial Times, Li claimed that while DeepSeek’s large language models had gained early traction, they were now limited by their text-only capabilities. He criticised the models for a higher rate of so-called “hallucinations”, as well as slower performance and greater cost compared to rival domestic offerings.

 

DeepSeek rapid rise in AI ecosystem

DeepSeek had rapidly risen to prominence in China’s artificial intelligence (AI) ecosystem. The company surged ahead in January with the release of its R1 model, widely seen as a significant leap in reasoning capabilities, and has since become the dominant player in the country’s large language model (LLM) space.

 

Competition heats up among AI firms

While Baidu was the first to respond to the release of OpenAI’s ChatGPT in 2022 with its Erniebot, DeepSeek and others have since entered the competitive landscape. Baidu was ultimately forced to open-source its models after subscription uptake fell flat, amid a wave of free alternatives.

 

In response to DeepSeek’s dominance, rivals have been racing to catch up. Baidu’s newly launched Ernie 4.5 Turbo and X1 Turbo feature multimodal capabilities — spanning text, image, audio and video — in a bid to outpace DeepSeek’s more narrowly focused approach. Alibaba has also made aggressive moves in the space, releasing its open-source QwQ model, which it claims matches DeepSeek’s performance with far less training data.