China may launch Mythos-class AI models before Q1 2027, Zai chief says after GLM 5.2 release

The US government recently banned Mythos 5 and Fable 5 for all foreigners, reportedly due to concerns of China-linked groups exploiting these models. However, Chinese AI lab Zai's chief has hinted that China may have its own Mythos-class AI model within a few months.

In Short

• China may launch its own Mythos-class model soon, says Zai chief
• Zai’s GLM 5.2 AI model already better than Google Gemini 3.5 Flash, Claude Sonnet 4.6
• US reportedly banned Mythos and Fable access to foreigners over China concerns

Artificial intelligence (AI) has become the new field of geopolitical tensions. The US has been trying to limit China’s ability to make or use advanced AI models for some time now. After banning sales of cutting-edge AI chips to China, the White House decided to restrict access of Anthropic’s Mythos 5 and Fable 5 AI models to all foreigners. As per reports, the US was concerned that China-linked groups may jailbreak these models for cyberattacks. However, Zai’s founder Tang Jie says that China may have its own Mythos-class AI model before 2027.

On X, a user discussed the possibility of Zai building a Mythos-class model following the release of GLM 5.2. The user claimed that Zai may have its own AI model on par with Mythos before the end of this year. The person wrote, “This means full PRC (People’s Republic of China) Mythos ("Fable") by Nov-Dec'26.”

Tech billionaire Elon Musk replied to this post, predicting that this may happen by next year. Musk said, “Probably Q1 (of 2027). However, Tang Jie claimed that Musk was likely incorrect. The Zai chief replied, “Won’t take that long.”

To give you some context on why this matters. Anthropic's Mythos-class models are claimed to be so powerful that they can hack any software out there. The Mythos model is said to have found tens of thousands of bugs already, which had not been previously not been identified. In the wrong hands, a model with such capabilities could potentially be used for cyberattacks.

If Tang Jie’s prediction comes true, US’ ban on Mythos and Fable to prevent China from using such models may not matter in the long run, as it may have its own models with similar capabilities.. Though it is likely that Anthropic and OpenAI will also make improvements to their own models in the same timeframe.

What is GLM 5.2?

GLM 5.2 is Zai’s latest AI model. On the Artificial Analysis Intelligence Index, it is now the leading open weights model, higher than other Chinese models like Kimi K2.6. GLM 5.2 scores 51 on the index.

For an open weight AI model, the model weights – the data of what the AI has learnt – are made publicly available, allowing anyone to download and run the model locally.

As a matter of fact, GLM 5.2 is so good that it even outperforms Google’s Gemini 3.5 Flash (50 points) and Anthropic’s Claude Sonnet 4.6 (47 points) models in the benchmark. Though it is behind Fable 5 (60 points) and OpenAI’s GPT 5.5 (55 points).

GLM-5.2 is the same size as Zai’s previous model, GLM-5.1, with 744 billion total parameters and 40 billion active parameters. However, the newer model scores 11 points higher in the index.

Another notable aspect of GLM-5.2 is that it was trained entirely on Huawei Ascend chips, with no Nvidia hardware involved. As per Stability AI founder Emad Mostaque, total training costs of this model was likely around $25 million, far cheaper than the cost it takes for Anthropic or OpenAI to train AI models.

Keep in mind that in the past, US AI companies have accused Chinese firms of distilling their AI models. Distillation refers to training a smaller model on the outputs of a larger one, so that it mimics the larger model's behaviour.

- Ends