Chinese technology giant Baidu on Friday launched two new AI models designed to offer enhanced features at substantially lower costs than comparable offerings from rivals like DeepSeek.
Speaking at a major developer conference in Wuhan, Baidu Co-founder, Chairman, and CEO Robin Li Yanhong unveiled the Ernie 4.5 Turbo and X1 Turbo models. Both boast advanced multimodal reasoning capabilities, allowing them to process and understand information from various formats like text and images.
Crucially, Baidu is positioning these models as highly cost-effective alternatives for developers.
According to Li, the Ernie 4.5 Turbo is approximately 40% cheaper than DeepSeek’s V3 large language model (LLM). The X1 Turbo reasoning model undercuts DeepSeek’s R1 model even more dramatically, offered at just a quarter of the price.
“The essence of innovation is the lowering of cost,” Li said during his keynote address.
He said that reducing the financial barrier to entry for powerful AI models would empower developers to shift their focus from managing expenses to creating “‘more super interesting and useful’ AI applications.”
This strategy aligns with Li’s previous predictions of explosive growth in AI applications within China this year. He reiterated this view at the conference, stating, “There will be a lot of [AI] models, but what will rule in the future are the apps.”
Li believes pairing specific models with tailored applications will unlock greater value.
The launch comes just weeks after Baidu made its flagship Ernie Bot service free for all users on April 1st, signaling an aggressive push for wider adoption amidst increasing competition from established players and innovative newcomers like Hangzhou-based DeepSeek.
While Baidu was the first major Chinese tech firm to release a ChatGPT-like LLM back in March 2023, its early lead has been consistently challenged.
Underscoring its commitment to fostering an AI application ecosystem, Baidu also launched the third iteration of its AI apps competition at the Wuhan event, pledging up to 70 million yuan (approx. US$9.6 million) in funding for winning developers.
Li framed the competition simply as a way to “help entrepreneurs build AI apps” and announced a goal to help cultivate 10 million AI talents over the next five years.
The conference itself highlighted Baidu’s central role in China’s AI landscape, attracting not only thousands of developers but also major international tech partners like Intel and Nvidia, alongside domestic companies including robotics start-ups Unitree Robotics and AgiBot, and EV makers Voyah and Leapmotor.
Li also showcased how AI integration is already enhancing Baidu’s existing product suite.
He noted significant growth in monthly active users for the AI-powered versions of Baidu Wenku (online documents) and NetDisk (cloud storage), reaching approximately 97 million and 80 million users, respectively.
Furthermore, he highlighted the Baidu Huiboxing platform, which can convert short human-hosted live streams into extended sessions led by virtual hosts, a tool reportedly adopted by some exporters impacted by US tariffs to boost domestic e-commerce efforts.
In a move reflecting broader trends in China’s tech sector, Baidu is also investing in domestic hardware capabilities.
Li confirmed the operational launch of an AI server cluster powered by 30,000 processors designed by Kunlun, Baidu’s own AI chip unit.
This cluster is capable of training large-scale AI models with parameter counts ranging from tens to hundreds of billions, strengthening Baidu’s infrastructure independence.
