Top 10 Best AI Large Model Companies in China

[baoshi.rao]

After ChatGPT sparked a wave of enthusiasm in China, the country's generative AI technology has also flourished.

The prospects for domestic AI models in China are very promising, especially under the impetus of national strategies, the field of artificial intelligence is developing rapidly. Chinese companies and research institutions are actively engaged in R&D and have made some significant progress.

Below, we take a look at the domestic large language models that have already begun internal testing.

1. Alibaba - Tongyi Qianwen

On April 7, Alibaba Cloud showcased their research achievements in the field of large models, announcing the start of user testing for their self-developed large model "Tongyi Qianwen." This marks another major Chinese company launching an AI large model for testing, following Baidu's "Wenxin Yiyan."

The official website for "Tongyi Qianwen" is http://tongyialyun.com. Upon opening the site, you can see its self-introduction: "I am Tongyi Qianwen, a large model specifically designed for human needs. I serve humanity, striving to make life better. I am an efficient assistant and a creative generation machine."

2. Baidu - Wenxin Yiyan

Baidu introduced a new generation of knowledge-enhanced large language model called "Wenxin Yiyan," a new member of the Wenxin model family. This model has the capability to interact with humans through dialogue, answering questions, assisting in creative tasks, and efficiently helping people access information, knowledge, and inspiration.

"Wenxin Yiyan" is developed based on Baidu's deep learning platform PaddlePaddle and the Wenxin knowledge-enhanced large model. It continuously learns from massive data and large-scale knowledge, featuring knowledge enhancement, retrieval enhancement, and dialogue enhancement technologies.

As one of the earliest large language model products to undergo internal testing in China, "Wenxin Yiyan" benefits from Baidu's accumulated experience and technical expertise since its full-scale commitment to AI in 2016.

3. Wudao - BAA's "Wudao" Large Model

The Beijing Academy of Artificial Intelligence (BAA) launched Wudao 2.0 in June 2021, China's first ultra-large-scale intelligent model system and one of the largest AI models in the world. Wudao is a language model designed to surpass OpenAI's GPT-3 and Google's LaMDA in human-level thinking. The model was trained on 4.9TB of images and text and outperformed state-of-the-art (SOTA) levels in nine benchmarks. Wudao is closer to achieving artificial general intelligence (AGI) and human-level thinking than any of its peers.

4. WeChat - WeLM

WeChat's AI product is called WeLM (WeChat Language Model). According to the official website, WeLM offers text continuation functionality but lacks native dialogue capabilities. It is a general-purpose language model highly skilled in understanding and generating text. By calling WeLM's API, various text-related tasks can be addressed. It's important to note that WeLM is not Tencent's Hunyuan large language model; Tencent has remained relatively low-key in the ChatGPT-like domain, with no related public announcements.

WeChat AI official website: WeLM

5. 360 - 360 Zhinao

360 Zhinao is a large language model developed by 360 Company. It utilizes deep learning technology to help enterprises better understand user needs and improve user experience.

The five core capabilities of 360 ZhiNao are perception, learning, reasoning, decision-making, and prediction.

It can help enterprises achieve intelligent management, thereby improving work efficiency and reducing costs.

Beta test address: 360 ZhiNao x 360 Search invites you to experience

6. SenseTime - SenseChat

SenseChat is a Chinese language model application platform under SenseTime.

SenseChat supports multi-turn dialogues and ultra-long text comprehension capabilities.

Features include:

• Programming assistant to help developers write and debug code more efficiently
• Health consultation assistant providing personalized medical advice
• PDF document reading assistant that can easily extract and summarize information from complex documents

Beta test address: https://chat.sensetime.com/wb/

7. Kunlun Wanwei - TianGong

TianGong is a large language model independently developed by Kunlun Wanwei and Singularity AI.

As a large language model, TianGong has powerful natural language processing and intelligent interaction capabilities, enabling various application scenarios such as intelligent Q&A, chat interaction, and text generation. It possesses rich knowledge reserves covering science, technology, culture, art, history and other fields. This model was jointly developed by Kunlun Wanwei and Singularity AI, and is China's first dual-100-billion-parameter large language model comparable to ChatGPT.

It can meet needs such as copywriting creation, knowledge Q&A, code programming, logical deduction, and mathematical calculation.

Beta test address: TianGong Large Model Official Website

8. Tsinghua - OpenBMB

A large-scale pre-trained language model library and related tools initiated by Tsinghua University team.

OpenBMB (Open lab for Big Model Base) aims to build a large-scale pre-trained language model library and related tools, accelerating the training, fine-tuning and inference of large models. It lowers the threshold for using large models and works with domestic and international developers to form an open-source community for large models, promoting the development of the large model ecosystem and realizing the standardization, popularization and practical application of large models.

The OpenBMB open-source community is jointly supported by Tsinghua University's Natural Language Processing Laboratory and the Language Model Acceleration Technology Innovation Center of Zhiyuan Research Institute. The initiating team has profound research foundations in natural language processing and pre-trained models, having published dozens of high-level papers at top international conferences in recent years on model pre-training, prompt fine-tuning, model compression techniques, etc.

Beta test address: OpenBMB - Bringing Large Models to Every Household

9. iFlytek - Spark Cognitive Large Model

iFlytek Spark is iFlytek's large language model.

The Spark Cognitive Large Model is a ChatGPT-like product released by iFlytek on May 6, 2023 at the Spark Cognitive Large Model results conference. As a new generation cognitive intelligence large model with Chinese as its core, it has cross-domain knowledge and language understanding capabilities. It can understand and execute tasks based on natural dialogue methods, continuously evolving from massive data and large-scale knowledge to achieve a closed-loop process from problem proposing, planning to solving.

10. Zhifou AI Q&A: Text-to-text + Multimodal AI Model ZhiFou AI Q&A is built on next-generation artificial intelligence AI knowledge-enhanced large language model technology, supporting text-to-text and multimodal generative AI capabilities, including copywriting generation, interactive dialogue, and intelligent Q&A. The official website is https://chat.cloudwizdom.com. The platform aggregates the most advanced machine learning models across various fields, providing one-stop services and dedicated customer support to help users better utilize large model technology.

Additionally, ZhiFou AI Q&A has released multiple product iterations. In May, it introduced multi-scenario features such as "Scenario Navigation" and "General Scenarios," with upcoming multimodal functionalities like "AI Image Generation" and "AI Video Generation." The model can transform text into videos. It consists of three sub-modules: text feature extraction, text-to-video latent space diffusion model, and video latent space to visual space conversion. The entire model contains approximately 2 billion parameters, with the diffusion architecture adopting the latest LLM framework. Through reverse diffusion processes, the model achieves video generation capabilities.

Domestic large language models also include Tencent's Hunyuan, NetEase's Yuyan, and JD.com's ChatJD. Although these models have not yet been publicly released or entered beta testing, they are under active development and will be rolled out gradually.

The prospects for domestic AI models are indeed vast. With China's national strategy driving advancements in artificial intelligence, the field is rapidly evolving. Chinese companies and research institutions are actively engaged in developing AI models and have made significant progress.

The advantages of domestic AI models include:

1. Talent Pool: China boasts a vast reservoir of tech talent, particularly in AI, which will fuel continuous innovation and development in domestic AI models.

2. Resource Advantages: China excels in computer hardware, cloud computing, and big data, providing essential resources for AI model development and training.

3. Market Opportunities: With its massive population and market, China offers extensive opportunities for the application of domestic AI models.

4. Government Support: Investments and policy support from the Chinese government will accelerate the growth of domestic enterprises and research institutions in AI model development.

In summary, the future of domestic AI models is highly promising. China's technological prowess and resource advantages will drive further progress and broader applications of AI models. AI is likely to become the next economic boom—if you missed the short-video wave, don't miss the AI wave!