Chinese AI models narrow the gap with US rivals

China has moved from being up to five years behind the United States in top-tier AI to trailing by just a few months, helped by a strategy focused on open models and cost efficiency. Chinese platforms including DeepSeek, Alibaba’s Qwen and Xiaomi’s MiMo are being deployed across businesses, government services and consumer devices despite US limits on high-end chips.

Moonshot AI’s Kimi K3 has become the latest signal of the narrowing gap, with benchmarks showing it can rival Anthropic’s Claude Fable 5 while using less computing power and costing far less. Alvin W. Graylin said Kimi K3 “commoditizes frontier models,” arguing that “For 98% of customers, the 1 to 2% difference [in knowledge] isn’t worth paying 10 to 20 times more.”

Washington is responding with threats to block US access to Chinese AI models and potential sanctions over alleged intellectual property theft through distilling. Anthropic and OpenAI accuse Chinese labs of harvesting advanced reasoning, while Beijing rejects the claims as “technological bullying.” Analysts say US policy is likely to harden as Chinese labs pursue public listings and investors weigh whether China’s lower-cost approach can compete with highly valued US AI firms.