New version of Baidu’s Ernie Bot delivers impressive results in testing

Baidu highlights significant enhancements in performance and functionality compared to its previous version, Ernie 3.0, and plans to integrate Ernie Bot into its broader ecosystem.


Baidu, the Chinese tech giant, has announced the latest version of its AI foundation model, Ernie 3.5, claiming superior performance compared to OpenAI’s ChatGPT in certain areas. The company cited a test conducted by China Science Daily, a local newspaper, which demonstrated Ernie 3.5 outperforming ChatGPT based on GPT-3.5 in general abilities.

Additionally, Baidu stated that Ernie 3.5 surpassed GPT-4 in various Chinese language capabilities. The testing involved the use of datasets such as AGIEval and C-Eval, which are benchmarks for evaluating AI model performance. Baidu highlighted significant improvements in efficacy, functionality, and performance compared to its earlier version, Ernie 3.0, with a notable boost in training and inference throughput.

Ernie Bot was introduced by Baidu to compete with OpenAI’s ChatGPT and Google’s Bard, and it has undergone public beta testing since its release. Baidu’s efforts in AI align with its significant investments in the field, including the creation of PaddlePaddle, China’s first open-source deep-learning platform. The company has also expressed its intention to integrate Ernie Bot into its wider ecosystem, including smart speakers and other devices.