中国人工智能公司 DeepSeek 在其服务出现近五小时的中断后，向用户道歉并承诺给予赔偿。
Chinese artificial intelligence firm DeepSeek apologised and promised compensation to users after its services suffered a nearly five-hour outage on Thursday.
根据这家总部位于杭州的初创公司在其应用程序编程接口（API）平台上发布的公告，DeepSeek 将为用户在 10 月 8 日至 10 月 14 日期间产生的所有费用全额退款，退款将以积分的形式发放给用户。
DeepSeek will issue users a full refund for any consumption between October 8 and October 14 in the form of credits, according to a notice posted on the Hangzhou-based start-up’s application programming interface platform.
DeepSeek 的同名聊天机器人以及其 V4 Pro 和 V4.1 Flash 模型的 API 服务在周四上午停止运行，导致全球用户受到影响；直到下午 1 点左右（即四小时五十分钟后），这些服务才恢复正常。
DeepSeek’s namesake chatbot and API services for its V4 Pro and V4.1 Flash models went offline on Thursday morning, affecting users globally, before being restored over four hours and 50 minutes later at around 1pm.
一些用户通过社交媒体分享了他们在尝试使用 DeepSeek 时遇到的错误信息，例如“服务器暂时不可用”。
Some users turned to social media to share the error messages they had encountered when trying to use DeepSeek, such as “server temporarily unavailable”.
DeepSeek 在公告中表示：“我们立即展开了全面调查，并采取了彻底的整改措施来确保服务的稳定性。我们还全面审查并消除了可能导致类似问题的潜在风险因素，以防止此类事件再次发生。”不过，该公司并未详细说明此次故障的具体原因。
“We immediately launched a comprehensive investigation and thorough rectification effort focused on service stability, and have comprehensively reviewed and eliminated single points of dependency to resolutely prevent similar issues from recurring,” DeepSeek said in the notice, without elaborating on the cause of the incident.
“我们对此给用户带来的不便和困扰表示诚挚的歉意。”此次故障发生之际，该公司正面临日益增长的基础设施压力——随着计算需求的激增，DeepSeek 需要扩大其团队规模和计算基础设施。据彭博社报道，该公司计划在中国内蒙自治区新增约 1 吉瓦（Gigawatt）的计算能力。
“We sincerely apologise for the inconvenience and disruption this may have caused.”The incident came as the company faces mounting infrastructure pressure amid surging compute demand. DeepSeek is looking to expand its teams and computing infrastructure as competition intensifies, with a plan to add about 1 gigawatt of computing capacity in China’s northern Inner Mongolia autonomous region, according to a Bloomberg report.
虽然 DeepSeek 之前也经历过多次服务中断，但这次中断是近几个月来最严重的一次。该平台曾在 3 月份也出现过一次持续约 12 小时的重大故障，影响了数亿用户。
DeepSeek has faced recurring service disruptions, but the latest outage was one of the most severe it has experienced in recent months. The platform also suffered a major disruption in March that lasted for around 12 hours, affecting hundreds of millions of users.
去年1月底，就在这家当时还鲜为人知的实验室发布了其R1推理模型并宣布将其免费向全球用户开放后，DeepSeek遭到了分布式拒绝服务攻击（DDoS攻击）。这些攻击的目的是通过大量网络流量来压垮该公司的服务器和带宽。该公司当时将这些攻击描述为“大规模的恶意攻击”。
In late January last year, shortly after the then little-known lab released its R1 reasoning model and made it free globally, DeepSeek was hit by denial-of-service attacks meant to overwhelm its servers and bandwidth with a flood of internet traffic. The company described them as “large-scale malicious attacks” at the time.