On 10/5 Beijing time, according to Bloomberg Intelligence, over the past few months, as models from Chinese AI labs such as DeepSeek have continued to advance, the performance lead of US AI companies over their Chinese peers has sharply narrowed to a historic low, threatening America's technological dominance.
Robert Lea, senior analyst at Bloomberg Intelligence, wrote in a Monday report that after DeepSeek released V4.1 Flash in the 9th month (September), China's top models trailed their US competitors on benchmark scores by only 3%. That gap has narrowed from about 9% in the 5th month (May) and 15% earlier this year.
He said this sustained performance improvement means Chinese competitors will further expand their market share.
Bloomberg noted that the rise of Chinese model technology has been driven by deepening AI expertise and researchers' ability to optimize models for domestically produced hardware. These advances have also raised questions about whether US technology export restrictions are effective.
Lea said China's progress "further calls into question whether the US's technological dominance in AI can be sustained over the long term."
DeepSeek's V4.1 Flash ranked sixth in the LiveBench global rankings last month, making it the highest-ranked Chinese model since the startup's breakthrough with its reasoning model R1 in 2025. LiveBench scores AI models based on their responses to and analysis of questions, puzzles or tasks, a process similar to measuring human IQ.
DeepSeek's most recent LiveBench score was 81.1, below Anthropic's top score of 83.4. Lea said this means DeepSeek's models are now performing on a par with the leading AI systems of Anthropic and OpenAI. However, although the score gap between the two sides has narrowed to just 3%, only 3 of the top 15 models selected by LiveBench come from China.
Lea cautioned that such rankings change constantly, and that monetization may be difficult regardless of where a model ranks on a leaderboard.