Baidu's Ernie 3.5 beats ChatGPT on some metrics

Baidu, China’s leading search engine provider, said the latest iteration of its Ernie AI model had surpassed the widely popular ChatGPT on multiple key metrics.

Baidu said in a statement on Tuesday that Ernie 3.5, the latest version of its Ernie AI model, had surpassed “ChatGPT in comprehensive ability scores” and outperformed “GPT-4 in several Chinese capabilities”.

The Beijing-based company cited a test that the state newspaper China Science Daily ran using datasets including AGIEval and C-Eval, two benchmarks used to evaluate the performance of artificial intelligence (AI) models.

OpenAI did not immediately reply to an email seeking comment outside of usual business hours.

The Baidu announcement comes amid a global AI buzz kicked off by ChatGPT that has spread to China, prompting a flurry of domestic companies to announce rival products.

Baidu was the first big Chinese tech company to debut a AI product to rival ChatGPT when it unveiled its language AI Ernie Bot in March. Ernie Bot, built on Baidu’s older Ernie 3.0 AI model, has been in invite-only tests for the past three months.

Other big Chinese tech firms, including Alibaba Group (9988.HK) and Tencent Holdings, have all since revealed their respective AI models.

Baidu said its new model comes with better training and inference efficiency, which positions it for faster and cheaper iterations in the future. Baidu also said its new model would support external “plugins”.

Plugins are additional applications that will allow Baidu’s AI to work on more specific scenarios such as summarizing long text and generating more precise answers.

ChatGPT rolled out plugin support in March.