Baidu, the Chinese technology giant, has launched two new artificial intelligence (AI) models, claiming they outperform competitors like DeepSeek and OpenAI on key benchmarks. As the race to dominate the large language model (LLM) space intensifies, Baidu has positioned its latest offerings as significant advancements in AI capabilities.
The company introduced Ernie 4.5, a multimodal foundational model, and Ernie X1, its first multimodal reasoning model, making both freely accessible on its website from Sunday. According to Baidu, Ernie 4.5 surpasses OpenAIโs GPT-4o on various benchmark platforms, including CCBench and OCRBench, in image, audio, and video processing areas. According to the company’s statement on WeChat, Ernie 4.5 outperformed DeepSeek V3 and was on par with OpenAIโs GPT-4.5 in text-based assessments.
Competing in Chinaโs AI market
Baidu was the first major Chinese tech firm to launch an LLM in March 2023, following the global excitement generated by OpenAIโs ChatGPT. However, its early lead has been challenged by domestic rivals, including Alibaba, Tencent, and ByteDance, which have rapidly gained traction among businesses and consumers. DeepSeek, in particular, has fuelled an open-source movement in China, forcing companies like Baidu to rethink their strategies to remain competitive.
While Baidu has not yet released benchmark results for Ernie X1, it claims the model delivers performance comparable to DeepSeek R1 at only half the cost. The company has priced business access to Ernie X1โs application programming interface (API) at 2 yuan (US$0.28) per million token inputs and 8 yuan (US$1.10) per million token outputs. In comparison, DeepSeek charges US$0.55 per million token inputs and US$2.19 per million token outputs for its DeepSeek-reasoner, which runs on the R1 reasoning model. Last month, DeepSeek raised its API prices due to rising demand.
A shift toward open-source AI
In a surprising move, Baiduโs founder, chairman, and CEO, Robin Li Yanhong, announced that Ernie 4.5 would become open-source from June 30. This marks a significant shift from his previous stance favouring closed-source AI development.
โOne thing we learned from DeepSeek is that open-sourcing the best models can greatly help adoption,โ Li said in a February earnings call. โWhen the model is open source, people naturally want to try it out of curiosity, which helps drive broader adoption.โ
Despite Baiduโs advancements in AI, the company is grappling with declining revenue. It reported a 2% year-on-year drop in total revenue for the fourth quarter, with full-year revenue down 1%. Weak advertising revenue continues to challenge the Beijing-based tech firm as it pushes forward in the competitive AI industry.
