China has released its first four politically approved large generative AI models, marking a new stage in how the country is organizing the development of AI-generated content (AIGC). The models come from Alibaba, Baidu, Tencent, and 360 Group, and each has passed China’s official "large model standard compliance assessment."
The assessment is not only a technical checkpoint. It is also part of a broader effort to create a compliance directory for the industry, giving companies a clearer route for releasing large models within China’s regulatory framework.
What China Approved
The four approved systems are large generative AI models from Alibaba, Baidu, Tencent, and 360 Group. The source names two of them directly: Baidu's ERNIE bot and Alibaba's Tongyi Qianwen.
Baidu's ERNIE bot has reached about 100 million users since its launch in August. Alibaba's Tongyi Qianwen is described as the only open-source model in the first group of released large-scale AI models.
The assessment was carried out by the China Electronics Standardization Institute, which operates under the Ministry of Industry and Information Technology, according to the Global Times. The stated purpose is to support the development of AI-generated content in industry and establish a compliance directory.
That directory matters because it turns model approval into a visible benchmark. For AI companies, passing the review can signal that a model is aligned with the official expectations now shaping the market. For customers and developers, it offers a clearer distinction between models that have completed the process and those that have not.
How The Review Works
The evaluation is based on a multidimensional model evaluation framework and an indicator system. The framework is designed to cover three broad areas: generalization, intelligence, and safety.
The review also spans multiple modalities. It includes text, speech, and visual content, which reflects how large generative AI models are no longer limited to written answers. A model may need to handle prompts, spoken input, generated responses, and visual material while still meeting the same compliance expectations.
The source does not provide the full scoring method or the internal indicators used in the assessment. What is clear is that China is treating large model evaluation as a structured process rather than a purely informal judgment. The combination of technical performance and safety review places model release inside a formal approval path.
For companies building AI services, that changes the operating environment. Model performance remains important, but it is not the only issue. A system must also be able to fit the compliance expectations attached to AIGC deployment.
Why Upfront Regulation Matters
Pan Helin, a professor at the International Business School of Zhejiang University, told the Global Times that government regulation upfront is better for the industry's growth in China than intervening later. That view frames the assessment as a way to guide the market early, rather than responding after models are already widely deployed.
The logic is straightforward. Generative AI systems can affect many forms of content, including text, speech, and visual output. If regulators want the industry to develop within defined boundaries, reviewing models before or during release gives companies clearer expectations.
The source also states that the Chinese Communist Party requires generative AI services to be based on the "core values of socialism." That requirement gives the compliance review a political dimension as well as a technical one.
In addition to evaluating and validating language models, China has released a dataset of 50 billion tokens in 100 million data points to train language models that reflect its political views. This shows that the regulatory approach is not limited to approval after training. It also reaches into the data environment used to shape model behavior.
The Performance Race
The approval process is unfolding while Chinese AI companies are also trying to narrow the performance gap with leading global systems. According to Dou Dejing, associate professor at Tsinghua University's School of Electronic Engineering, China's best domestic AI models have reached the level of GPT-3.5, and the technical gap with GPT-4 is narrowing.
Baidu CEO Robin Li described his company's Ernie Bot 4.0 as being on par with OpenAI's GPT-4 in late October 2023. Zhou Hongyi, founder and president of 360 Security Technology, sees ChatGPT 12 to 18 months ahead of the Chinese competition, according to the Global Times.
Those assessments point to a competitive landscape with two simultaneous pressures. Chinese companies are working to improve model capability, while also demonstrating that their systems can satisfy official requirements. The first four approved models therefore represent more than product releases. They show how model development, market access, safety evaluation, and political alignment are becoming linked in China’s AI sector.
For the broader AI industry, China’s approach highlights a central question: how should powerful generative systems be evaluated before they reach users? In China, the answer now includes a formal compliance assessment, a directory-building effort, and explicit political requirements. The first four approved models give that system its first visible test cases.