彭博情报：中美AI模型性能差距缩至3%，美国领先优势降至历史最低
彭博情报（Bloomberg Intelligence）高级分析师 Robert Lea 周一发布的报告指出，中国顶尖AI模型与美国最强模型的性能差距已缩小到仅3%，降至历史最低：年初时这个差距约为15%，5月份约为9%。 这波收窄主要来自深度求索（DeepSeek）9月发布的 V4.1 Flash 模型。它在 LiveBench 测评上拿到81.1分，排名全球第六，成为迄今排名最高的中国模型，与榜首 Anthropic 的83.4分已经十分接近。 报告认为，中国AI的追赶得益于技术积累加深，以及针对国产硬件的优化调优，中国团队的低成本路线正不断蚕食美国大模型的市场份额。但报告也提醒，基准排名变化很快，且中国AI行业仍面临盈利难、千余个大模型恶性价格战的问题。（资料来源：彭博情报报告；彭博法律、Aju Press 等媒体转引）
Bloomberg Intelligence: US-China AI Performance Gap Narrows to Record-Low 3%
Top Chinese AI models now trail their leading U.S. rivals by just 3% on benchmark scores, the narrowest gap on record, Bloomberg Intelligence senior analyst Robert Lea wrote in a report published Monday. The gap stood at about 9% in May and around 15% earlier this year. The narrowing was driven by DeepSeek's V4.1 Flash, released in September. It scored 81.1 on LiveBench, ranking sixth globally, the highest ever for a Chinese model, and within striking distance of the top score of 83.4 from Anthropic. The report attributes China's ascent to deepening AI expertise and researchers' ability to optimize models for domestic hardware, warning that the trend points to further market share gains for Chinese labs. But it also cautions that benchmark rankings can change quickly and that China's AI industry faces an unresolved profitability problem, with more than 1,100 large language models competing in a domestic market marked by aggressive pricing. (Sources: Bloomberg Intelligence report, via Bloomberg Law, Aju Press and others.)
图片来源：图片来源：Openverse · Policy Exchange（CC BY）