Overview of Chinese Companies' ChatGPT-like Products and Large Language Model Initiatives

On March 16, Baidu released its generative AI product "Wenxin Yiyan" and opened testing, followed by 360's demonstration on March 29, Alibaba's "Tongyi Qianyu" internal testing invitation on April 7, and other major releases in April and May.

1. Baidu – Wenxin Yiyan supports literature creation, copywriting, mathematical computation, and multimodal generation. It is built on Baidu's self-developed AI infrastructure, featuring the ERNIE 3.0 Zeus model with billions of parameters, the PaddlePaddle deep learning platform with a unified static‑dynamic framework and heterogeneous parameter server, and the Kunlun 2 AI chip (256 TOPS@INT8, 128 TFLOPS@FP16).

Wenxin Yiyan quickly attracted many users, and Baidu also offers the enterprise‑level model service platform "Wenxin Qianfan" that includes the full suite of Wenxin models and development toolchains.

2. Tencent – Hunyuan announced on March 30 that it is developing a ChatGPT‑style chatbot for integration into QQ, WeChat, and Tencent Cloud. Tencent's Hunyuan series covers NLP, CV, and multimodal models, and has achieved top rankings on CLUE benchmarks. It is already used in WeChat search and advertising, and a dedicated "Hunyuan Aide" project is led by Dr. Zhang Zhengyou.

3. Alibaba – Tongyi Qianyu opened internal testing invitations on April 7. The product builds on Alibaba's "Tongyi" large model series and the OFA unified learning paradigm, supporting over 30 multimodal tasks, including image description, visual grounding, and text generation.

4. ByteDance has formed separate language and vision model teams, with the language team led by the search department and the vision team by the intelligent creation department. Their AI products, such as the video generation tool Jianying, are in early stages.

5. Huawei – Pangu announced the upcoming release of Pangu 4 on April 8, featuring NLP, CV, and scientific computing models. While Pangu has fewer parameters (1.2 trillion) than GPT‑3, it leverages extensive Chinese text data to excel in Chinese language tasks.

6. NetEase – Ziyue introduced the education‑focused ChatGPT‑style model "Ziyue", leveraging NetEase's long‑standing language learning data and upgraded NPU hardware to improve inference speed and accuracy.

7. 360 – Zhi Nao opened reservations for its "Zhi Nao" model on April 9, with testing starting April 16. The model demonstrates reasoning abilities but still requires multiple prompts for satisfactory answers.

The article also discusses the broader challenges faced by Chinese ChatGPT equivalents, including incomplete data corpora, multilingual limitations, compliance and content moderation requirements, and the need for transformative application scenarios beyond simple integration.

Industry experts anticipate that the first to achieve a breakthrough in AIGC will set the standard for future AI development in China.