Baidu Launches Low-Cost ERNIE 4.5 Turbo & X1 Turbo Multimodal AI Models

At its annual developer conference in Wuhan, Baidu CEO Robin Li introduced upgraded versions of its foundational multimodal model ERNIE 4.5 Turbo and inference model ERNIE X1 Turbo, highlighting enhanced multimodal capabilities, strong inference, low cost, and free access via Ernie Bot.

Li emphasized that the models are designed to help developers build optimal applications without worrying about model capability costs or tooling, noting that advanced AI chips and massive language models have little value without real‑world use cases.

According to Baidu’s press release, the new products “break the limits of multimodal and inference models,” and ERNIE X1’s performance is comparable to DeepSeek R1 while costing only half as much.

The company plans to integrate the two models into its ecosystem, including the Baidu Search engine and other services.

Li also announced that Baidu has successfully built a cluster of 30,000 self‑developed third‑generation P800 chips capable of training models similar to DeepSeek.

Paul Smith‑Goodson, VP of Moor Insights & Strategy, said the P800 cluster indicates Baidu is ready to train models with billions of parameters, but its chip performance still lags behind U.S. GPUs, and without benchmark data his assessment remains cautious.

He added that the development reflects a broader U.S.–China race to create the first artificial general intelligence (AGI) model, with the United States still leading.

Thomas Randall, AI market director at Info‑Tech Research Group, noted that Baidu remains a key player in China’s competitive AI landscape, alongside Alibaba, Tencent and Huawei, and that its ERNIE series is one of the few domestic LLMs comparable to OpenAI’s GPT‑level models.

Baidu’s New Large Language Model Holds Strategic Significance

Overall, analysts view Baidu’s advancements as strategically important globally, though its commercial impact in Western markets remains limited; the race to innovate in AI is no longer U.S.–centric.