As China’s AI race accelerates, ‘model fatigue’ becomes the next challenge

On September 22, the global artificial intelligence landscape experienced a remarkable shift in a matter of hours. In Beijing, Xiaomi engineers streamed the training run of their new MiMo-V2.6 model, while Anthropic unveiled its frontier Opus 5.5 model from San Francisco. Shortly after, OpenAI surprised the industry with the release of its GPT-6 Sol and Luna models.

While Silicon Valley grappled with what was dubbed "model fatigue", China's tech sector faced the phenomenon on a much larger scale. Analyst Rui Ma noted that the rapid pace of model releases in China was generating a more intense version of "model fatigue" than in the United States. The economic pressure in China fell more heavily on developers, as deeply embedded models in consumer apps and cloud platforms triggered brutal price wars.

Companies like DeepSeek, Z.ai, Tencent, Alibaba, and Moonshot AI were racing to release new models at a breakneck pace, with some slashing prices to remain competitive. The model layer in China proved to be the least sticky layer, with users more price-sensitive and brand loyalty difficult to build. Despite calls for a slowdown from industry leaders, Chinese developers saw no option but to keep up with the rapid releases.

Analyst Poe Zhao emphasized that businesses in China prioritize stability, predictable output, integration costs, data security, and technical support over incremental improvements in benchmark scores. To survive the churn, Chinese tech firms are carving out niches, such as ByteDance focusing on video generation and Ant Group concentrating on healthcare.

The competition remains fierce, with Alibaba reporting triple-digit growth in AI-related product revenue.