News
The three major overseas AI collectives were down for nearly 4 hours. Zhipu posted "We are still online" and enabled free nighttime services. The leading AI services in the United States collectively experienced failures. ChatGPT, Claude, and Grok announced service abnormalities. Google Gemini and Microsoft Copilot also received a large number of interruption reports.
3 min read
Source: Telegram AI频道
The three major overseas AI collectives were down for nearly 4 hours. Zhipu posted "We are still online" and started free nighttime services. The top AI services in the United States collectively experienced failures. ChatGPT, Claude, and Grok announced service abnormalities. Google Gemini and Microsoft Copilot also received a large number of interruption reports. This large-scale outage lasted for about 3 hours and 40 minutes, and the fault first broke out around 9:23 a.m. ET on September 3. Claude was the first to report errors, and OpenAI received more than 37,000 fault reports at the peak. According to data from the overseas fault monitoring platform Down Detector, Anthropic's Claude series models were the first to experience a surge in error rates, covering core models such as Mythos 5.1, Opus, and Fable 5.1. Just two minutes later, xAI's Grok service was completely offline; about an hour and a half later, OpenAI's ChatGPT and the programming tool Codex also experienced large-scale access errors. During the peak period of the incident, there were more than 37,000 fault reports for the OpenAI service worldwide, 80% of which were concentrated on ChatGPT. During the same period, Google Gemini and Microsoft Copilot received a large number of user outage feedbacks, and the AI programming tool Cursor directly announced that some of its services were interrupted due to upstream large model failures. As of 11:16 noon Eastern Time on the same day, the main services of Claude were the first to resume, and then ChatGPT and Grok gradually returned to normal operations. Domestic large models are strong, and Zhipu is taking advantage of the trend to launch nighttime free services. It is worth noting that during the collective failure of overseas large models, domestic large models remained stable. That night, Zhipu posted on social media X: "We are still up." - translated as "We are still online." At the same time, Zhipu also announced on the evening of September 3 that from now until September 20, from 23:00 every night to 9:00 the next day, GLM-5.3-Flash in the calling package can be used for free in Z Code. The usage quota in other agents supported by the package will be doubled, and it will automatically take effect after entering the specified time period. According to reports, GLM-5.3-Flash is the first native multi-modal model of the GLM-5 series. It scored 57 points in the world's authoritative Artificial Analysis comprehensive intelligence index, entering the global cutting-edge model capability range. This model is suitable for code generation, code