China Will Have Mythos-Class AI Before 2027, Zai Founder Declares After GLM-5.2 Release
A public debate over China's progress toward frontier AI capabilities has erupted on social media, with Zai founder Tang Jie pushing back against Elon Musk's timeline projections. The exchange follows the release of GLM-5.2, Zai's newest flagship model, which has demonstrated performance approaching that of Anthropic's most advanced systems .
When an X user asked about China's timeline to achieve "Fable-class" capabilities—referring to Anthropic's recently released frontier models—Musk responded that such a milestone would likely occur in the first quarter of 2027 . Tang Jie's response was brief but definitive: China "won't take that long" . This bold assertion signals growing confidence in China's AI ecosystem, built on the capabilities demonstrated by GLM-5.2.
GLM-5.2: The Model That Changes the Equation
The debate is grounded in real technical achievement. GLM-5.2, a 744-billion-parameter Mixture-of-Experts model with 40 billion active parameters, has emerged as the leading open-weight model on the Artificial Analysis Intelligence Index with a score of 51 . It outperforms Google's Gemini 3.5 Flash (50) and Anthropic's Claude Sonnet 4.6 (47), though it remains behind Anthropic's Fable 5 (60) and OpenAI's GPT-5.5 (55) .
On key benchmarks, the model is closing the gap:
FrontierSWE, a benchmark for autonomous software development, GLM-5.2 scored 74.4—just 0.7 points behind Anthropic's Claude Opus 4.8 and ahead of GPT-5.5 at 72.6 .
SWE-bench Pro, which measures the ability to resolve real-world GitHub issues, the model scored 62.1, surpassing both GPT-5.5 at 58.6 and its predecessor GLM-5.1 at 58.4 .
The model features a 1 million-token stable context window—five times larger than GLM-5.1's 200,000-token limit—enabling complex agentic tasks across entire software repositories .
Trained Without Nvidia, Fully Open Source
Perhaps the most significant aspect of GLM-5.2 is its training infrastructure. Zai has confirmed that the model was trained entirely using Huawei Ascend chips, with no Nvidia hardware used in the process . This represents a breakthrough in China's ability to develop world-class AI without relying on American semiconductor technology, amid ongoing US export restrictions.
Stability AI founder Emad Mostaque estimated training costs at approximately $25 million—significantly less than what Anthropic or OpenAI spend on comparable frontier models . The model is available under the MIT license, allowing unrestricted global access without country-based restrictions .
The US Export Ban and Its Implications
The GLM-5.2 launch came against a backdrop of escalating US-China AI tensions. On June 13, the US Commerce Department ordered Anthropic to suspend all foreign access to its Fable 5 and Mythos 5 models, citing national security concerns over a potential jailbreak technique . The order covered users outside the US and even foreign-national Anthropic employees. Anthropic was forced to disable both models for all users to ensure compliance .
Anthropic disputed the order's justification, arguing the technique was narrow, affected only one specific scenario, and produced similar outputs from other publicly available models including GPT-5.5. The company warned that applying this standard across the industry "would essentially halt all new model deployments" .
The Strategic Significance
The combination of GLM-5.2's open-source availability and strong performance creates an interesting dynamic. As Zai founder Tang Jie noted in a widely shared post following the US export ban: "At the moment when frontier models are being cut off without reason, we are more convinced than ever: science should be global. The path to AGI must not be walled off"
If Tang Jie's prediction proves accurate, the US restrictions on Anthropic's most advanced models may have limited long-term impact on China's AI capabilities. The race to achieve Mythos-class AI is accelerating, and China's timeline may be shorter than many expect.
