The performance gap between top Chinese and U.S. AI models has narrowed to 3%, hitting a record low

Odaily News: As Chinese AI labs such as DeepSeek continue to make progress, the performance lead of top U.S. AI models over Chinese models has narrowed to about 3%, down from about 9% in May this year and 15% at the beginning of the year, reaching a record low.

DeepSeek V4.1 Flash, released in September, ranked sixth globally on LiveBench with a score of 81.1, approaching Anthropic's highest score of 83.4. Robert Lea, a senior analyst at Bloomberg Intelligence, said that Chinese AI models continue to improve in specialized capabilities and optimization for domestic hardware, but commercialization still faces pressure from low profit margins, price competition, and competition among a large number of models. (Bloomberg)