ChatGPT, Codex error rate skyrockets! OpenAI’s 22 services are simultaneously downgraded, with 8 major incidents in 5 days

OpenAI’s official status page posted a new incident at 5:17 PM Taiwan time on July 25, with the title Elevated error rates (error rates increased). The official page left only one line: “We are investigating,” without detailing the cause or the estimated time to recovery. There are 22 components marked as performance degradation; besides ChatGPT and Codex, the list also includes Login, Conversations, Files, Search, the Sora, and VS Code extension components. This is the eighth incident since July 21.
(Background: OpenAI admitted that its AI model accidentally infiltrated Hugging Face)
(Additional context: Notion temporarily interrupted Claude model connection; product lead: it’s just a temporary service failure)

Key takeaways

• OpenAI’s status page at 5:17 PM on 7/25 posted “Elevated error rates”; the status is still under investigation
• A total of 22 services were marked as degraded performance, and the overall classification is Partial System Degradation
• Looking through the incident list, from July 21 to now the official has posted eight incidents; the first seven have all been resolved

OpenAI’s official status page posted a new incident at 5:17 PM Taiwan time on July 25, titled Elevated error rates, meaning error rates increased. It can be used, but the probability of things going wrong may be higher.

The official notes on the investigating list mention problems with these services. No cause is given, and no recovery time is provided. As of 5:50 PM, the incident status is still “investigating,” already more than half an hour.

Not only ChatGPT and Codex are included in the list. The status page marks a total of 22 components as degraded performance, including Login, Conversations, Responses, Embeddings, Files, Search, Batch, Fine-tuning, Moderations, Sora, GPTs, Codex API, VS Code extensions, ChatGPT Atlas, and ChatGPT Work. The overall status is marked as Partial System Degradation.

Eighth incident within five days

This incident’s impact is classified by the official as minor. But when you look back at the incident list, the picture isn’t quite the same.

• 7/21: Image generation was completely unavailable in ChatGPT; classification major
• 7/22: File uploads and image generation errors, and on the same day Workspace Agent errors
• 7/23: Three incidents in one day, including elevated error rates for ChatGPT and API errors and latency for gpt-image-2
• 7/24: Codex Review errors increased
• 7/25: This incident

Five days, eight incidents—seven have been marked resolved, and the eighth is still under investigation. OpenAI has not explained whether these incidents are related.

Frequently asked questions

Is ChatGPT down right now?

Starting at 5:17 PM Taiwan time on July 25, OpenAI’s official status page reported elevated error rates. Twenty-two services including ChatGPT and Codex are marked as performance degradation. This is partial degradation rather than a full shutdown. As of the time of this report, the status page still shows investigating.

Which services are affected by this outage?

In addition to ChatGPT and Codex, it also includes Login, Conversations, Responses, Embeddings, Files, Search, Batch, Fine-tuning, Sora, GPTs, Codex API, VS Code extensions, ChatGPT Atlas, and ChatGPT Work—both the developer side and the consumer side are affected.