DeepSeek V4.1 Flash takes full advantage, and the gap between the LiveBench scores of top models in China and the United States narrows to 3%DeepSeek V4.1 Flash 发力，中美顶尖模型 LiveBench 跑分差距缩至 3%

Beijing time on October 5, according to Bloomberg Industry Research, in the past few months, as the models of Chinese AI laboratories such as DeepSeek have continued to improve, the performance lead of U.S. AI companies relative to their Chinese counterparts has shrunk sharply to an all-time low, putting the U.S.’s technological dominance under threat. After DeepSeek released V4.1 Flash in September, top Chinese models trailed U.S. rivals by just 3% in benchmark scores, Robert Lea, senior analyst at DeepSeek Bloomberg Intelligence, wrote in a report on Monday. The gap narrowed from about 9% in May and 15% earlier this year. This continued improvement in performance means Chinese competitors will further expand their market share, he said. Bloomberg pointed out that the rise of China’s model technology is due to the continuous deepening of its AI professional capabilities and the ability of researchers to optimize models for domestic hardware. These developments also raise questions about the effectiveness of U.S. restrictions on technology exports. Leah said China's progress "further calls into question the long-term sustainability of U.S. technological dominance in AI." DeepSeek's V4.1 Flash ranked sixth in the LiveBench global rankings last month, becoming the highest-ranked Chinese model since the startup's breakthrough in 2025 with its inference model R1. LiveBench scores an AI model's responses and analysis of a question, puzzle, or task, a process similar to measuring human IQ. DeepSeek’s most recent score on LiveBench was 81.1, lower than Anthropic’s highest score of 83.4. Leah said this means the DeepSeek model's performance is already "comparable to leading AI systems from Anthropic and OpenAI." However, although the score gap between the two sides has narrowed to just 3%, only three of the top 15 models selected by LiveBench are from China. Leah cautions that such rankings are constantly changing, and monetization can be difficult regardless of a model's position on the leaderboard.