ChatGPT was down again – here's everything we know about the latest outage

When did the problems start?

According to Downdetector, the latest ChatGPT outage started at around 9am ET / 2pm BST, when we saw a big spike in reports in both the US and UK.

The issues are seemingly intermittent, with the main problems including "something went wrong" error messages in ChatGPT's browser version and "upstream connect errors" in the mobile app.

We've seen both issues on the TechRadar team, though it doesn't appear to be a complete outage that's affecting everyone equally...

You may be seeing this a lot today...

It isn't yet clear what's causing today's ChatGPT problems, but OpenAI's status page says it's investigating a live incident.

The problems seem to be intermittent and producing a variety of lovely symptoms, including the "something went wrong" message above and the classic "upstream connect error" (always a favorite) on the mobile app.

While it's a bit too soon to blame the Studio Ghibli image craze, it's hard to ignore the timing of Sam Altman's increasingly panicked X posts telling us to stop making so many...

OpenAI acknowledges ChatGPT issue

So it wasn't just our collective imaginations – OpenAI's Status page is now showing an "Increased Error Rates" issue for ChatGPT.

The notice states that "we are investigating the issue" but doesn't provide any more information about a cause, other than it's been "ongoing for 2hr".

The description seems to fit our experience – while it isn't a complete outage for the service, we have certainly been seeing a lot or error messages.

The Downdetector reports seemed to have dropped slightly from their earlier peak, but not by far – with the number hovering at around the 1,000 mark at the timing of writing. Seems like ChatGPT had a heavy weekend on the old Ghibli cocktails...

Is Gemini having issues too?

Maybe we do need more of those AI chips and GPUs after all – it seems Google Gemini also experienced a minor outage at the same time as ChatGPT, though its issues seems to be largely resolved already.

We noticed error messages like the above in Gemini recently and Downdetector showed a fairly big spike in reported problems at around 9.36am ET / 2.36pm BST. This could point to DNS issues affecting both Gemini or ChatGPT, or it may be unrelated.

Either way, Gemini seems to be back to its old self – while OpenAI is seemingly having a second spike in reported problems, according to Downdetector...

Is "biblical" demand to blame?

While OpenAI hasn't officially commented on what's causing today's ChatGPT problems, it's hard to escape the conclusion that the soaring demand for its new image generator has at least had at least some impact.

Sam Altman has been posting increasingly concerned messages to X (formerly Twitter) over the past few days, starting by stating that "our GPUs are melting" from the Ghibli craze on March 28 to a "this is insane our team needs to sleep" yesterday on March 30.

In a reply, Altman added that OpenAI hadn't been able to catch up with the increased load, adding "biblical demand, I have never seen anything like it". Well, perhaps OpenAI should have thought of that when it stoked the Ghibli fire.

In our tests, we've found that ChatGPT's image generator is still working, but is very slow going – much like the rest of the AI chatbot today...

OpenAI may be rolling out a fix...

Some potential good news for those who are struggling to get ChatGPT to make images of monkeys riding Segways (or help with more serious work) – the OpenAI Status page now says "we are working on implementing a mitigation".

Yes, that statement is so non-committal is must have been written by ChatGPT, but it at least promises that the AI chatbot will be running a bit more smoothly soon. The spike in Downdetector reports have also dropped significantly, so perhaps it's already working for some...

A promising sign...

OpenAI has now removed the message acknowledging "increased error rates" on ChatGPT from its status page – which suggests the problems are now easing.

That's also reflected in the Downdetector graph above too, which shows that only around 120 people reported problems with OpenAI in the US at the time of writing (and that number is fortunately even lower in the UK).

This means today's issues likely won't be a repeat of the big outage in January 2025, which saw over 10,000 people report problems at a peak that was caused by a widespread "bad gateway error". Good news – you may recommence your Ghibli-themed image generation, unless you're too much of a fan to do it...

ChatGPT is back to its old self

Aside from an ongoing wrinkle around new signups for ChatGPT and Sora in some regions, the AI chatbot now seems to have recovered from its latest wobble – and is back to full working order.

That's good, because its Monday 'to do' list contains around two billion requests for Studio Ghibli-style images and more, so it had better get cracking.

We'll be keeping a close eye on ChatGPT this week, because it's clearly been under strain from some unpredented load following the roll-out of its viral image generation feature. But for now, we're happy to see that it's back to its error message-free self.

The ChatGPT problems return...

At almost exactly the same time as yesterday (9am ET / 2pm BST), there's been another spike in reports of ChatGPT problems on Downdetector – and OpenAI has also confirmed "login issues" on its Status page.

At the time of writing, there have been almost 1,500 reports from frustrated ChatGPT users in the US on Downdetector, though the spike in the UK has almost completely reversed.

The good news is that the 'history' section of OpenAI's status page posted at 9.20am ET / 2.20am BST that "all impacted services have now fully recovered" and that "users globally should be able to login without issue". So hopefully this tremor should be short-lived and a bit less disruptive than yesterday's...

Looks like ChatGPT had two issues

OpenAI posted on its Status Monitor that its login issues were resolved at 9.20am ET / 2.20pm BST, so why have lots of users been having problems since? It turns out a second issue, our old friend "increased error rates" from yesterday, has also returned over the past few hours.

That might explain why ChatGPT has been uncooperative for you again today. Fortunately, at 11.12am ET / 4.12PM BST, OpenAI posted that "all impact services have now fully recovered", so we should see those error messages fizzle out, leaving us to melt OpenAI's GPUs again in peace.

OpenAI CEO Sam Altman responds

We now have our first comment about ChatGPT's patchy performance this week from OpenAI CEO Sam Altman (above). His post on X (formerly Twitter) confirms what most of us have been thinking – that the problems are related to the company's frantic rollout of major features like ChatGPT's new image generator.

Yesterday Altman was crowing that ChatGPT had acquired "one million users in the last hour". To put that in perspective, ChatGPT took five days to attract the same number of users when it launched in 2022 – when it became the fastest-growing consumer app in history.

So yes, we can see why OpenAI might be experiencing capacity issues, while also having lots of sympathy for the paying ChatGPT Plus users who are wondering why on earth the company just rolled out the new image generation feature to free users – and is now apparently considering doing the same with Deep Research.

'One step at a time' clearly isn't really a phrase in OpenAI's lexicon, for better and worse...

The ChatGPT problems return – and Sora joins the party

ChatGPT is again experiencing problems, which started at around the same time as yesterday – a little after 9am ET / 2pm BST. The issues, acknowledged by OpenAI as "increased error rates", has this time been joined by "degraded performance" for the video-generating Sora.

If the past two days are any guide, the problems shouldn't last long – and OpenAI says it's now "working on implementing a mitigation", which has previously ironed out the wrinkles in performance pretty quickly.

This all undoubtedly relates to Sam Altman's warning that "you should expect new releases from OpenAI to be delayed, stuff to break and for service to sometimes be slow" as it deals with "capacity challenges".

In other words, OpenAI's been biting off a bit more than it can chew by handing out new features like confetti. Short-term performance hits sound reasonable for free users, but it's a bit more galling for paid ChatGPT Plus users like me...

This is becoming a familiar sight...

I can understand the odd blip on ChatGPT when it rolls out new features, but this week's issues hopefully aren't a sign of things to come. It feels like the quality of ChatGPT Plus is being compromised by OpenAI's need to roll out new features on its free tier at a rate of knots, before capacity is ready.

Competition from the likes of DeepSeek – which we've described as smarter, cheaper, faster and a real rival to OpenAI's models – is hot, but OpenAI needs to be careful it doesn't take its paying subscribers for granted as it scales and rolls out its new treats.

We're now hearing that Deep Research could be coming to OpenAI's free tier soon. Maybe sort out these "capacity challenges" first eh, OpenAI?

The biggest outage of the week (so far)

Okay, today's ChatGPT outage has now mushroomed into the worst one of the week so far, according to Downdetector reports (above).

In the US, reports peaked at almost 2,200 in the US at 11.10pm ET / 4.10pm BST. Fortunately, they're now down to around half that. And in the UK, there were around 300 reports at the peak, also now down to around half of that figure.

In theory, the "degraded performance" issues should be starting to ease – OpenAI's Status page says that it's "applied the mitigation" and is "monitoring the recovery". Let's hope those graphs keep falling in the right direction...

And we're back (hopefully)

OpenAI's "mitigation" seems to be doing the trick – ChatGPT is now working fine for me, and reports on Downdetector in the US and UK have now dropped back to baseline.

The question now is whether or not this is going to become a daily occurrence this week. For three days running, the problems have started at around the same time – about 9am ET / 2pm BST. If you've got something urgent to get done at those times tomorrow, you may want to work around those times.

Understandably, ChatGPT users – particularly paying subscribers – are getting a bit miffed about the continued problems this week. Let's hope we're not all back here tomorrow looking at Downdetector...