A new, inexpensive Chinese AI model is catching up with Anthropic, OpenAI
Chinese AI startup DeepSquare released a new model, DS-4, claiming performance comparable to Anthropic's Claude and OpenAI's GPT-4 at a fraction of the cost. The model's low training and inference expenses highlight China's continued ability to produce competitive AI systems despite US chip export restrictions, raising concerns about the erosion of Western tech advantages.
Background
- The article refers to a new Chinese AI model that reportedly achieves performance competitive with leading U.S. frontier models (such as those from Anthropic's Claude or OpenAI's GPT series) at a fraction of the cost.
- This follows a trend where Chinese AI labs — like DeepSeek, Alibaba's Qwen, or Baidu's ERNIE — have been releasing increasingly capable models that rival Western counterparts, often with lower training budgets.
- The "catching up" narrative matters because it challenges the assumption that cutting-edge AI requires the massive compute budgets of U.S. tech giants; cheaper models could accelerate global AI adoption and intensify geopolitical competition over AI standards, export controls (e.g., U.S. chip restrictions), and talent.
- Prior context: DeepSeek's 2024 release shocked the industry by matching GPT-4-class performance with reportedly far fewer GPUs; this new model appears to continue that trend.