A little more than a year ago a small Chinese artificial-intelligence startup shocked the world(震惊世界). When DeepSeek unveiled(推出，发布) a pair of models that performed nearly as well as the best Western ones, but were built for a fraction of the cost(一小部分成本), a panic(恐慌) swiftly followed. The share prices of Nvidia and other providers of AI infrastructure briefly tumbled(暴跌) as investors fretted(焦虑，担忧) (wrongly, it turned out) that demand for their wares would slow in the face of such a leap(飞跃，巨大进步) in the efficiency of model-making. Yet the release on April 24th of the lab’s new model, called v4, has been greeted with a shrug(反应冷淡，耸肩以对). Why?
大约一年多前，一家小型中国人工智能初创公司震惊了世界。当DeepSeek推出两款性能几乎与西方最佳模型相当、但成本仅为其一小部分的模型时，恐慌迅速蔓延。英伟达及其他AI基础设施供应商的股价短暂下跌，因为投资者担心（事实证明是错误的）在模型制造效率如此飞跃的情况下，对其产品的需求将放缓。然而，该实验室于4月24日发布的新模型v4却遭到了冷遇。为什么？
DeepSeek’s latest release hits many of the same heights(达到同样高度) its predecessor did. According to tests run by the company, the performance of its most powerful “Pro” system falls only marginally short of(仅略低于) the models put out by leading American competitors three to six months ago. DeepSeek’s v4 is cheap for customers, too. An introductory offer(促销价格，尝鲜价) makes it a thousandth of the price of the best American models for some uses. Even after that rate expires(到期，失效) on May 7th, v4 will cost between a tenth and a quarter(十分之一到四分之一之间) of American equivalents.
DeepSeek的最新版本达到了与上一代相同的许多高度。根据该公司运行的测试，其最强大的“Pro”系统的性能仅比美国领先竞争对手三到六个月前推出的模型略逊一筹。DeepSeek的v4对客户也很便宜。一项促销活动使其在某些用途上的价格仅为美国最佳模型的千分之一。即使该价格于5月7日到期后，v4的成本仍将相当于美国同类产品的十分之一到四分之一。
But it seems that, unlike DeepSeek’s previous blockbuster(重磅产品，热门货), v4 was not particularly cheap to build. In 2025 the lab eagerly(急切地，热切地) pointed out that the cost of training its AI was about $6m, far below the going rate(现行价格，通行费率) in the West. The lab’s technical white paper on v4, however, omits(省略，遗漏) any estimate of this measure. The fact that 16 months elapsed(流逝，过去) between v4 and its predecessor also hints that(暗示……) oodles(大量，很多) of processing power were used to train it.
但看来，与DeepSeek之前的热门产品不同，v4的构建成本并不特别便宜。2025年，该实验室热切指出其AI训练成本约为600万美元，远低于西方的普遍价格。然而，该实验室关于v4的技术白皮书却省略了这一成本的任何估算。v4与其前代产品之间相隔16个月，这一事实也暗示着它使用了大量的计算能力进行训练。