How to access Chinese LLM chatbots across the world

Accessible Chinese LLM Chatbots 

Since the summer of 2023, hundreds of Chinese large language models have become available to the public, thanks to the government allowing AI companies to open their models for general use. For users in the West, navigating these models can be challenging due to language barriers and registration requirements, such as needing a valid Chinese phone number. However, many of these chatbots support English and are surprisingly accessible. Whether you’re curious about their performance or want to conduct serious experiments for work, there are numerous ways to explore Chinese LLM-powered chatbots.

If your phone has Chinese number 

If you have a Chinese phone number, accessing these models is a breeze. In China, a domestic phone number often serves as a proxy for identity verification. This means you can easily access almost any online service, including AI chatbots, by visiting the model's website and using your phone number to register an account.

In our tests, we discovered that while some large language models are accessible to users outside China, a few major ones—including those developed by Huawei, the cybersecurity firm 360, and the Chinese Academy of Sciences—require a Chinese phone number for access. Unfortunately, obtaining a local phone number outside China can be nearly impossible. Fortunately, there are alternative methods to explore these models.

Access without a Chinese Number 

You can access without a Chinese number

A few Chinese AI platforms provide direct access to their chatbots for users outside China. Simply sign up, enter your phone number, and verify with the code sent via text message.

For instance, Doubao, an AI developed by ByteDance, can generate text and images, search the internet, and summarize uploaded documents. ChatGLM, a multimodal tool by the emerging Chinese AI unicorn Zhipu, offers similar features and is also accessible with non-Chinese phone numbers.

Additionally, DeepSeek, a relatively new AI company, stands out for its affordability. It offers free access for casual users, while business users pay only $0.14 per million tokens, compared to OpenAI’s GPT-4 Turbo at $10 per million tokens. You can test its text and coding chatbots on their website by registering with an email address.

Hugging Face 

Hugging Face is one of the largest global AI communities today, functioning like a GitHub for AI. Many open-source large language models (LLMs) have shared their code and applications on Hugging Face’s platform.

Chinese companies have also joined this trend, making their code available for AI developers. Some even offer demos on the site, allowing non-technical users to easily test the models without any coding. Since Hugging Face doesn’t require a Chinese phone number for registration, it provides a convenient way to explore many Chinese models.

Another Chinese tech giant, Tencent, has also contributed to Hugging Face by uploading its AI model, Hunyuan-DiT, which generates images from text prompts. However, other models based on the Hunyuan foundation are not available on the site.

ModelScope: A Gateway to Chinese AI Models

In November 2022, Alibaba launched ModelScope, a platform resembling a domestic version of Hugging Face. It serves as a hub for the Chinese AI community to gather, access open-source models, and engage in discussions. Notably, ModelScope allows users to register with non-Chinese phone numbers, providing another effective workaround for accessing Chinese LLMs. Companies like Baichuan, a promising AI startup backed by Alibaba, have made their models available on ModelScope for anyone to test.

One of the most useful features on ModelScope is the "LLM Arena" application, created by Open Compass, an AI lab in Shanghai that evaluates different models. As the name suggests, the arena lets users compare two models side by side by having them respond to the same prompt. This application integrates 11 Chinese models, making it easy for non-Chinese users to access several AI tools that are otherwise restricted.