• China’s AI race heats up as Baidu claims its new Ernie X1 matches DeepSeek R1 at half the price.
• Chinese tech giants close capability gap with Western AI leaders on price and open-source approach.

China is advancing in the AI race at an accelerating rate. The most recent development is tech giant Baidu’s launch of two new AI models that it claims outperform offerings from domestic challenger DeepSeek and Western leader OpenAI in several benchmark tests.

On March 16, 2025, Baidu made its latest multimodal foundational model, Ernie 4.5, and its first multimodal reasoning model, Ernie X1, available on Ernie Bot’s official website. The Beijing-based company’s announcement marks an escalation of the competition between Chinese AI developers, who are now challenging one other’s achievements and, increasingly, positioning themselves as alternatives to Western AI powerhouses.

Benchmark battles: Chinese models take on global leaders

According to a statement from Baidu on WeChat, Ernie 4.5’s multimodal capabilities – covering images, audio, and video – outperformed OpenAI’s GPT-4o on benchmark platforms, including CCBench and OCRBench.

The company further claimed that, based on several benchmarks, Ernie 4.5’s text capabilities surpassed those of DeepSeek V3 and “roughly matched” OpenAI’s GPT-4.5. Perhaps most notably, Baidu asserts that its new reasoning model Ernie X1 “delivers performance on par with DeepSeek R1 at only half the price” – a strategic pricing move that could accelerate X1’s adoption in price-sensitive markets.

Reasoning models represent an evolution in AI development, as unlike standard large language models that excel at generating conversational text, reasoning models are designed to handle complex problem-solving tasks that require logical thinking, planning, and use of online tools.

DeepSeek’s R1, launched in late 2024, has become the new standard for AI model makers to beat on price and, arguably, performance. This was particularly impressive, considering the company spent only a fraction of what larger competitors invest in training models.

Pricing has become a key battleground in the industry. Baidu has set business access to Ernie X1’s API at 2 yuan (US$0.28) per million token inputs and 8 yuan per million token outputs. That significantly undercuts DeepSeek, which charges US$0.55 per million token inputs and US$2.19 per million token outputs for its DeepSeek-reasoner service powered by the R1 model. The Hangzhou-based startup recently raised its API prices amid surging demand, creating an opening for Baidu’s competitive pricing strategy.

From closed source to open: A strategic reversal

The shift toward open-source development represents another pivotal change in China’s AI landscape. Baidu’s founder, chairman, and CEO, Robin Li Yanhong, announced last month that Ernie 4.5 would become open source on June 30, a dramatic reversal from his previous advocacy for closed-source AI development. “One thing we learned from DeepSeek is that open-sourcing the best models can greatly help adoption,” Li said during an earnings call with analysts in February.

“When the model is open source, people naturally want to try it out of curiosity, which helps drive broader adoption,” he said.

The pivot toward open-source development suggests a new phase in the competitive dynamics of the Chinese AI market. Competitive intensity in China’s AI market escalated in March 2023, when Baidu became the first Chinese Big Tech firm to launch an LLM following OpenAI’s ChatGPT release.

However, in the last two years, other Chinese AI players have challenged Baidu’s early-mover advantage. The South China Morning Post reported that DeepSeek has championed the open-source movement, while companies including Alibaba, Tencent, and ByteDance have been “galloping ahead in grabbing business and consumer users for their models,” the publication’s editorial read.

Despite technological advances, Baidu continues to face business challenges. The company reported that total revenue for the fourth quarter fell 2% year-on-year, with full-year revenue declining 1%. This indicates that even with cutting-edge AI capabilities, monetisation remains challenging, given the costs associated with training and running AI models.

The implications of China’s advancing AI capabilities extend beyond corporate competition. The global AI landscape reshaping as Chinese AI models achieve performance metrics comparable to their Western counterparts, and offer lower prices. Intense competition offers benefits for global AI users: more capable models at lower prices, with options spanning proprietary and open-source approaches.

The proliferation of high-performing models creates new opportunities for businesses building on AI yet complicates decisions about which AI infrastructure to adopt. As Baidu’s latest announcement demonstrates, China’s AI race is not merely about catching up to Western innovation – it’s increasingly about setting competitive terms through aggressive pricing, open-sourcing, and specialised capabilities like reasoning models.

The rapid development cycles and fierce competition among Chinese tech companies suggest that the pace of AI innovation will continue, with potentially significant implications for the global AI ecosystem.