China’s Baidu to release next-gen AI model in 2025

China’s Baidu reportedly plans to launch the next generation of its large language model in the third quarter of 2025. A source revealed that the AI model will have updates in its multimodal capabilities. 

According to a CNBC report, China’s Baidu is set to release the next generation of its artificial intelligence model in 2025. The company first launched its ChatGPT-like Chatbot in 2023. However, the product was eclipsed by other Chinese and US LLM models. 

China’s Baidu to release AI model update amid DeepSeek market disruptions 

China’s Baidu’s LLM model, Ernie, is set to receive an update in the second half of 2025. Ernie 5.0 is reportedly a foundational model that will have enhancements in its multimodal capabilities. Multimodal AI is said to process images, audio, and texts and convert them across different categories. 

The planned launch comes after China’s startup DeepSeek disrupted the AI industry last month after unveiling its open-source LLM model. Chinese companies raced to develop LLM models to compete with OpenAI and other US-based startups.  

DeepSeek is said to have appealed to users due to its reasoning capabilities while undercutting OpenAI’s ChatGPT deployment costs. Shortly after DeepSeek’s launch, the global tech market experienced stock sell-offs that disrupted markets. A Bernstein analyst commented the launch defied US sanctions by proving that China could compete globally using domestically developed AI hardware and software.

Shortly after the launch, Baidu announced that it would support DeepSeek by running the AI model on its own computing chips. Other companies, such as Alibaba, Tencent, and ByteDance, announced they would also make DeepSeek accessible through their cloud platforms. Lenovo and Geely added that they had adopted DeepSeek into their cloud platforms and products. 

Baidu CEO Robin Li commented at the World Governments Summit in Dubai this week that the tech industry is experiencing an exciting time. He added that the interference costs of foundational AI models could be reduced by over 90% in the next year. 

Li also suggested that reducing the costs of deploying LLM models by a certain percentage will increase productivity at the same rate. The CEO added that DeepSeek’s emergence proved the unpredictable nature of innovation. He, however, noted that investment in data centers and cloud infrastructure is still necessary despite DeepSeek challenging the deployment costs for large LLM models. 

Baidu reveals its Wenku Platform gained more users in 2024 

Baidu revealed in January that its Wenku platform, used to create presentations and documents, had attracted over 40 million paying users by December 2024, a 60% increase from the previous year. In January, the company unveiled updated features to its users, such as using AI to create presentations based on a company’s financial filings. 

The current Ernie model Gen 4 was released in 2023, while its upgraded version was released in August 2024. The model supports gen AI integration across Baidu’s products, including content creation. The company has not yet officially announced plans to launch the next-generation update.

OpenAI boss Sam Altman commented on DeepSeek’s launch on Monday, acknowledging the Chinese tech startup did some “nice work” in creating an OpenAI rival chatbot. The CEO added he had yet to meet the competing firm next Tuesday. Altman revealed he was not surprised by the launch but admitted he had not anticipated the exact timing of the ChatGPT’s rival launch. 

The OpenAI CEO recognized that regional differences in AI products were inevitable. He noted that the language models would likely operate differently in the countries due to the current geopolitics.

Altman, however, raised concerns about democratic AI usage, adding that it may have unprecedented effects on society. He said that he hoped states could agree on rules on how the technology would operate in different countries.  

The ChatGPT boss also discussed OpenAI’s latest innovation, deep research, which aims to conduct complex research on users’ behalf.

Cryptopolitan Academy: FREE Web3 Resume Cheat Sheet - Download Now