Baidu Inc. released a new artificial intelligence model that articulates its reasoning, in an apparent bid to regain momentum against up-and-coming rivals like DeepSeek.
The Ernie X1 model by China’s internet search leader works similarly to DeepSeek R1 — which shocked Silicon Valley by offering comparable performance to the world’s best chatbots at a fraction of their development cost. Baidu’s reasoning model excels in areas like daily dialogs, complex calculations and logical deduction, it said in a statement Sunday.
Baidu also upgraded its flagship foundation model to Ernie 4.5. It immediately made all tiers of its service — including the X1 model — free for its chatbot users, several weeks than earlier previously planned.
The Beijing-based company was the first in China’s trillion-dollar tech sector to launch a chatbot modeled after OpenAI’s ChatGPT, but rival chatbots from ByteDance Ltd. and Moonshot AI soon took over in popularity. Open-sourced models like Alibaba’s Qwen and then DeepSeek gained greater recognition within the global developer community.
Ernie 4.5 outperforms OpenAI’s latest GPT 4.5 in text generation, Baidu said, citing several industry benchmarks.
Baidu has declared that it will make Ernie AI models open-source from June 30, representing a major strategic shift after the rise of DeepSeek. It also integrated the R1 model into its search engine — its bread-and-butter business.
The generative AI boom showed up in Baidu’s December-quarter results via a 26% jump in cloud revenue. That rise, driven by services provided to developers chasing computing power, was overshadowed by weak advertising sales amid China’s economic malaise.
Baidu concluded last month a drawn-out deal to acquire the YY Live streaming platform Joyy Inc. The $2.1 billion takeover released some $1.6 billion that Baidu previously deposited into escrow accounts, which it plans to invest into AI and cloud infrastructure.