AI race heats up: Chinese models challenge US rivals with lower prices

DeepSeek-R1, a reasoning model developed by China’s leading AI start-up, scored 60 on the Artificial Analysis Intelligence Index, making it the third most intelligent AI model globally.

It trails OpenAI’s o1 and o3-mini, which scored 62 and 66, respectively. The index evaluates model performance across seven categories, including reasoning, knowledge, and math & coding.

In terms of pricing for developers, DeepSeek-R1 ranks eighth globally, while OpenAI's GPT-4.5 and o1 are the most expensive models. The rankings highlight how Chinese AI models are swiftly catching up to Western counterparts in both performance and pricing, amid a fierce domestic price war, as reported by the South China Morning Post.

"One year ago, the AI frontier was overwhelmingly dominated by U.S. companies," Artificial Analysis noted in a post on X last month. "Today, nearly a dozen Chinese companies have models matching or exceeding current-generation models from most U.S. labs."

DeepSeek's logo on a smartphone. Photo by VnExpress/Bao Lam

Alibaba Group Holding is also a prominent player, with its latest reasoning model, QwQ-32B, launched earlier this month. It ranked fourth in intelligence and tenth in pricing.

Both DeepSeek-R1 and QwQ-32B have been deemed more intelligent and less expensive than offerings from Western companies like Anthropic’s Claude 3.7 Sonnet, Mistral AI’s Mistral Large 2, and Amazon’s Nova Pro, according to the analysis.

DeepSeek-R1 charges $2.19 per million tokens of output for users accessing the service via its application programming interface.

In contrast, U.S. companies continue to rely on extensive computing resources to train their models, resulting in higher costs. OpenAI’s o1 charges users $60 per million tokens of output, nearly 30 times more than DeepSeek-R1.

On Thursday, OpenAI launched o1-pro, claiming it provides "consistently better responses" than the original o1 reasoning model. OpenAI is charging $150 per million tokens (~750,000 words) fed into the model and US$600 per million tokens generated, making it the company’s most expensive AI model yet, as reported by TechCrunch.

Alibaba’s cloud unit recently added DeepSeek models to its Bailian platform, offering one million free tokens per user for the full-scale V3 and R1 models. Distilled versions cost just 0.5 yuan (US$0.07) per million tokens, which it calls the "lowest price on the market."