OpenAI Core Services Suffer Major Outage

On July 25, OpenAI's core products—API, ChatGPT, and Codex—experienced a suspected simultaneous massive outage. Numerous users reported being unable to use the services normally, and the official status page showed significantly elevated error rates while the team investigated. Users later reported that services were gradually restored. This event had a broad impact on both the consumer side and the developer ecosystem.

Confirmed

According to screenshots of OpenAI's official status page, APIs, ChatGPT, and Codex were experiencing elevated error rates, and the official status was set to "investigating" (lasting for about 31 minutes). Multiple users confirmed the server-side unavailability in their posts: several Reddit users reported being unable to log into the OpenAI developer dashboard and receiving persistent primaryapiservererror when calling the API. On the consumer side, users found ChatGPT messages stuck in the "Sending" state or failing directly. Codex frequently showed 503 Service Unavailable and biscuitbakerservicemecircuitopen error codes. Additionally, users reported "Message delivery timed out" errors on the web client, which even caused a 16-hour running task to suddenly disappear. Multiple bloggers pointed out that this was not a localized network issue for a single user, but a severe service interruption covering the entire site. According to user feedback, the outage was relatively short-lived, and subsequent reports indicated that service appeared to have been restored.

Unconfirmed

There is no official conclusion yet regarding the specific cause of this outage. While complaining about Codex frequently throwing 503 errors, Twitter user @kimmonismus joked that GPT might have been "shaken" by the release of the competing model Opus 5, but this is merely community banter and not the actual cause.

Why it matters

With AI services deeply integrated into development and daily workflows today, a complete outage of core interfaces and products is highly destructive. As @CSProfKGD pointed out, the scope of an AI outage's impact can already be compared to traditional internet outages. The simultaneous failure of the API, ChatGPT, and Codex not only disrupted daily interactions for regular users but also directly paralyzed third-party developer applications relying on OpenAI's interfaces, once again highlighting the challenges of high availability in current AI infrastructure.

2026-07-25 ~ 2026-07-25 · 16 related posts

Full story(2 episodes)→

Primary sources

1 near-duplicate retellings: proofofbeef