AnswersPublished23 sources
ChatGPT outage on July 23: What happened and how widespread was it?
The July 23, 2026 incident was a degraded performance event—not a confirmed total global shutdown—with elevated errors affecting ChatGPT, Codex and OpenAI APIs. More than 300 user reports were recorded around the Thursday disruption, while reports described failed or stalled requests on ChatGPT’s web and mobile expe...
Published byEdited with GPT-5.6 LunaImages generated with GPT Image 1.5
Research answer
Key takeaways
- The July 23, 2026 incident was a degraded performance event—not a confirmed total global shutdown—with elevated errors affecting ChatGPT, Codex and OpenAI APIs.
- More than 300 user reports were recorded around the Thursday disruption, while reports described failed or stalled requests on ChatGPT’s web and mobile experiences.
- The incident formed part of a cluster of OpenAI disruptions in late July, including a broader July 25 outage and a more heavily reported ChatGPT disruption on April 20.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What happened during the Thursday ChatGPT outage, which services and users were affected across web and mobile platforms, how widespread wer. Article summary: The Thursday incident appears to refer to July 23, 2026: OpenAI reported elevated error rates affecting ChatGPT, Codex, and its API. It was a degraded-performance event rather than a confirmed total global shutdown, but . Topic tags: general, general web, education. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake nu
The Thursday outage was the July 23, 2026 disruption in which OpenAI reported elevated error rates affecting ChatGPT, Codex and its API. It was better described as degraded performance or a partial outage than as a confirmed worldwide shutdown: many users encountered failures, but the available evidence does not show that every account, plan or region was affected. 30
29
What happened during the July 23 outage?
Users reported that ChatGPT requests failed, stalled or could not load conversations. Reports also included problems using the web and mobile experiences, while Codex and OpenAI’s API were listed among the affected services. 22
29
32
OpenAI’s incident record described the problem as “elevated error rates.” That wording is important. It indicates that requests were failing at an unusually high rate, but it does not mean that the service was completely inaccessible to all users. 30
The University of Colorado’s IT alert, which tracked the same service disruption for its users, said OpenAI was working with a downstream infrastructure provider while attempting to restore normal service. It reported service recovery earlier on July 23, although OpenAI’s own incident record continued monitoring the event into the following day. 29
30
How widespread were the reported errors?
Downdetector-related reporting recorded more than 300 user reports shortly before 3 p.m. on July 23, with nearly 90% of the complaints concerning ChatGPT. 22
That figure should not be treated as a measure of total downtime or total failed requests. Outage trackers count people who submit reports, so the number can be influenced by awareness, geography and the visibility of the problem. The official status language—elevated error rates—provides a better description of the incident’s technical scope than the raw report count alone. 22
30
The available reports establish user impact across ChatGPT and related OpenAI services, but they do not establish that every user, subscription tier or geographic region experienced the same failure. That is why “partial outage” or “degraded performance” is more precise than “ChatGPT was completely down.”
What did OpenAI say about identifying and fixing the problem?
OpenAI’s status updates moved through the standard incident-response sequence:
- The company began by investigating the issue affecting the listed services.
- It identified elevated error rates and worked on a mitigation.
- It applied the mitigation and monitored recovery.
- The incident was ultimately marked resolved at 19:33 UTC on Friday, July 24, with impacted services reported as fully recovered.
30
The timeline was not identical across all reports. The University of Colorado notice said a fix had restored ChatGPT, OpenAI APIs and Codex at 3:36 p.m. on July 23, while the status-page record retained the incident for longer recovery monitoring. 29
30 This may reflect uneven recovery, different reporting checkpoints or OpenAI’s decision to keep monitoring after service became usable for many customers. The public material does not provide enough information to determine which explanation is definitive.
OpenAI also did not provide a detailed technical root cause in the available incident updates. The downstream-provider reference in the university alert is the clearest public indication of the infrastructure context, but it should not be presented as a complete causal explanation. 29
How did it compare with the July 25 outage?
The July 23 incident was part of a difficult stretch for OpenAI rather than an isolated disruption. On July 25, ChatGPT, Codex and OpenAI’s APIs again experienced elevated error rates. Users reported inaccessible conversations, stalled prompts and 503 errors, including an internal error label that prevented requests from reaching OpenAI’s servers. The Next Web described that event as the company’s fourth service disruption in four days. 7
Other reporting characterized the July 25 event as a broader global outage, with users in multiple countries unable to load conversations, access saved history or run prompts. 3
15
The distinction between the two dates matters: July 23 involved a substantial but not necessarily universal degradation, while July 25 generated reports of a wider and more visible failure across consumer and developer-facing services.
How did the July incidents compare with April?
The April 20 disruption was also more heavily reported. Moneycontrol cited more than 1,720 Downdetector reports involving gateway errors, login failures and an inability to load conversations, while TechRadar reported that OpenAI classified the event as a partial outage lasting at least 90 minutes. 11
12
Reporting on April 20 also described problems accessing projects and, in some cases, ongoing work. A mitigation was deployed around 1 p.m. Eastern time, after which recovery was monitored. 14
The clearest comparison is therefore one of pattern and visibility: April 20 was a larger, sharply reported ChatGPT disruption, while late July brought repeated incidents across ChatGPT, Codex and the API over several days. The available sources do not establish that all of these events had the same technical cause.
What did the outage mean for workplace users?
For individuals, the practical symptoms were failed prompts, loading problems and inaccessible conversations. For teams, the consequences could extend beyond the ChatGPT interface. Organizations using the API or Codex may have faced failed automation, delayed jobs, broken developer workflows or backlogs in systems that depend on OpenAI responses. Security and platform guidance following the July incidents specifically urged teams to check whether API-dependent systems produced silent failures, queued work or visible errors, and whether autonomous tasks stopped safely. 6
The exact business cost was not publicly quantified. There is no reliable evidence in the available sources for a dollar-loss estimate, and the operational impact would vary by organization, workload and fallback capability.
For teams that depend on OpenAI services, the incident highlights several practical reliability questions:
- Are failed requests retried with backoff rather than duplicated?
- Do applications distinguish a temporary provider error from a completed response?
- Are queued jobs observable when the API is degraded?
- Can users switch to a documented fallback when ChatGPT or Codex is unavailable?
- Do multi-step or autonomous workflows fail safely instead of stopping halfway through a task?
6
Why did prediction markets pay attention?
The repeated July disruptions also drew attention from prediction markets that define an outage using an official “partial” or “full outage” classification on a particular day. 18 Other markets tracked how many outage days ChatGPT might record during July.
20
Those markets are useful as a signal of public interest in AI-service reliability, but their prices are not evidence about the cause, severity or future likelihood of an outage. They reflect trading behavior and the wording of each contract, not an independent reliability measurement.
Bottom line
The July 23 ChatGPT incident was a real OpenAI service disruption affecting ChatGPT, Codex and the API, but the strongest evidence supports describing it as elevated error rates and degraded performance rather than a total global shutdown. More than 300 user reports captured the visible impact, while OpenAI’s status timeline moved from investigation to mitigation and monitoring before marking the incident resolved at 19:33 UTC on July 24. 22
30
Its larger significance was the surrounding pattern. The July 23 event was followed by another major disruption on July 25, while April 20 had already shown how a ChatGPT incident could affect conversations, projects and API-dependent work. For businesses using AI in production, the takeaway is less about one outage count than about building observability, retries, safe failure handling and practical fallbacks into the workflow.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "ChatGPT outage on July 23: What happened and how widespread was it?"?
The July 23, 2026 incident was a degraded performance event—not a confirmed total global shutdown—with elevated errors affecting ChatGPT, Codex and OpenAI APIs.
What are the key points to validate first?
The July 23, 2026 incident was a degraded performance event—not a confirmed total global shutdown—with elevated errors affecting ChatGPT, Codex and OpenAI APIs. More than 300 user reports were recorded around the Thursday disruption, while reports described failed or stalled requests on ChatGPT’s web and mobile experiences.
What should I do next in practice?
The incident formed part of a cluster of OpenAI disruptions in late July, including a broader July 25 outage and a more heavily reported ChatGPT disruption on April 20.