Ant Group boasts of breakthrough with new fast, cheap Chinese AI models
Ant Group, a digital payments company backed by Alibaba’s Jack Ma, just released two new AI models: Ling-Plus and Ling-Lite.
The models use a similar “mixture-of-experts” technique, like DeepSeek’s R1 models.
But the significant breakthrough is that these models were trained on Chinese-produced chips from Alibaba and Huawei Technologies. Most of the AI world has been using Nvidia’s GPUs for training, but tight US export controls have kept the most powerful chips out of the hands of Chinese AI companies (though some do still slip through).
Ant says its models match or beat leading models from DeepSeek, Mistral, and Meta’s Llama 3.1. This is another boon for China’s global AI ambitions, as it shows that its scrappy AI developers can innovate around attempts by the US to hamstring their progress in the field.
These models were trained for about $880,000, according to Ant, and “improve the efficiency and accessibility of AI development in resource-constrained settings,” its research paper says.
But the significant breakthrough is that these models were trained on Chinese-produced chips from Alibaba and Huawei Technologies. Most of the AI world has been using Nvidia’s GPUs for training, but tight US export controls have kept the most powerful chips out of the hands of Chinese AI companies (though some do still slip through).
Ant says its models match or beat leading models from DeepSeek, Mistral, and Meta’s Llama 3.1. This is another boon for China’s global AI ambitions, as it shows that its scrappy AI developers can innovate around attempts by the US to hamstring their progress in the field.
These models were trained for about $880,000, according to Ant, and “improve the efficiency and accessibility of AI development in resource-constrained settings,” its research paper says.
