What to use when ChatGPT is down

You are halfway through a task when the reply stops mid-sentence, or the page hangs.

Here is the fast version. Work out whether the problem is ChatGPT or your side. That takes about thirty seconds. Then move the task somewhere else while OpenAI fixes it.

Thirty seconds: is it ChatGPT, or is it you?

Run these in order. Stop as soon as one works.

Open a brand new chat and send "hello". If that answers, your old thread is the problem.
Reload the page. Then try the app if you were on web, or the web if you were on the app.
Turn off Wi-Fi and use mobile data. Plenty of "outages" are one network.
Try a private window. Extensions and stale sessions break more chats than servers do.
Check a status feed. Ours sits at is ChatGPT down, read from OpenAI's own feed.

If a fresh chat on a different network still fails, it is not you.

Four different things people mean by "down"

Each one needs a different move.

Full outage. Nothing loads at all. Rare, usually short, usually confirmed on a status page within minutes.

Degraded. It loads, but replies crawl, cut off, or come back wrong. This is the common one. It is also the one status pages are slowest to admit.

One model, not the whole service. The newest model errors while an older one from the same company answers fine. Providers ship and roll back individual models constantly.

Your account. You hit a limit, a payment failed, or a region restriction kicked in. The service is up. You are just not in it.

Why regenerating genuinely helps

That sounds like folk advice. There is a real mechanism behind it.

On 20 August 2026, Grok started replying in nonsense. Actual output, quoted by TechCrunch: "match it without and your they and two for planets can practical and often cheese." Grok's official account confirmed it the next morning, calling it "a rare temporary generation glitch" and telling people to start a fresh chat or regenerate. It cleared.

Glitches like that attach to a session or a single request. A new chat gets you a fresh roll. Try twice before you conclude anything.

Status pages lag. Plan around it.

status.openai.com is the official word. It is also written by the company having the bad day.

The pattern repeats at every provider. Users notice first. Downdetector spikes. The status page updates twenty minutes later, or never, because the incident never crossed an internal threshold.

Treat a green status page as weak evidence. If your requests are failing, your requests are failing.

Our own AI status monitor tries to close that gap two ways. It reads each provider's official feed every three minutes. It also flags a model separately when requests we actually sent in the last thirty minutes failed, which can happen while the provider still reports everything as normal.

Where to check, provider by provider

Bookmark the one you use. Each of these reads the vendor's own feed.

The fifth case: the feature was retired

Worth ruling out before you spend an hour debugging.

Features get retired. Microsoft removed five consumer Copilot features by 18 August 2026, Deep Research and the Mico character among them. Nothing was down. They had shipped the shutdown.

If one button stopped working and everything else is fine, search the product's changelog before you search for an outage.

What to use instead, right now

Pick by what you were doing.

Long writing or careful reasoning. Claude Sonnet 5.5, or Claude Opus 5.5 for the heavier work.
Big documents and fast answers. Gemini 3.7 Flash.
Anything about this week. Any model with web search switched on, such as Gemini 3.7 Flash or Claude Sonnet 5.5.
A second opinion on the same prompt. Run it in two models and read both.

None of them will sound exactly like ChatGPT. For most tasks that matters less than you expect. Claude, Gemini and Grok all have free tiers you can sign into in about a minute.

Take the work with you

Switching models is easy. Rebuilding the context is the annoying part.

Copy your last prompt before you close anything.
Paste in the two or three earlier messages that actually matter. Not the whole thread.
Add one line: "continue from here, same tone and format."
Leave the original tab open. Compare when it comes back.

Most models pick up a half-finished draft without complaint, as long as you hand them the draft.

Where MultiChats fits, honestly

We are one app with 60 models from 18 providers, OpenAI, Anthropic, Google, xAI and Mistral among them. When ChatGPT will not answer, you switch model inside the same thread and carry on. No new tab, no re-explaining.

Now the honest part. We are not immune to any of this. We call the same providers you do. When OpenAI has an outage, our GPT models go down with it. When Anthropic has one, Claude goes down here too.

What changes is the blast radius. One provider failing takes out a slice of the app instead of all of it. That is the entire pitch and we are not going to inflate it.

If we are the ones broken, is MultiChats down will say so.

Keep this bit

Fresh chat, different network. That is your thirty-second triage.
Regenerate twice before you believe the worst.
Green status pages lie by omission.
Have a second model you are already signed into, before you need it.
Copy the prompt before you close the tab.

Outages are usually short. Losing an hour to one is mostly a preparation problem.

Common questions

Is ChatGPT down right now?

Check ChatGPT status, which mirrors OpenAI's official feed. If it reads operational and you still cannot send, work through the thirty-second list above.

Will I lose my ChatGPT history during an outage?

Almost never. Threads live on OpenAI's servers and come back with the service. The message you were mid-way through is the part at risk, so copy it out.

What is the closest thing to GPT for the way it writes?

Most people land on Claude Sonnet 5.5 or Gemini 3.7 Flash. Send the same prompt to both and keep whichever reads better to you.

Try it on the free tier, or see what each plan covers under pricing.