Your AI assistant has gone quiet in the middle of a task. Is it you, or is it them?
This guide walks through the checks in order and explains what common error codes mean. It ends with what to do while a real outage lasts.
Step 1: Open the Official Status Page
Each provider runs its own status page and posts incidents there once it is aware of a problem.
| Product | Official status page | What it covers |
|---|---|---|
| ChatGPT and the OpenAI API | status.openai.com | The API, ChatGPT, Codex and other OpenAI products |
| Claude and the Claude API | status.claude.com | The Claude API, claude.ai and Claude Code |
| Gemini API and Google AI Studio | aistudio.google.com/status | The Gemini API and AI Studio |
| Gemini on Google Cloud (Vertex AI) | status.cloud.google.com | Google Cloud services, including Vertex AI |
Look at three things:
- The banner at the top. A banner such as all systems operational means the provider knows of no problem right now.
- The component list. Pages split the service into parts: the API, the chat app, logins, file uploads, a specific model. A problem with one part can leave the others working.
- Open incidents. A posted incident moves through stages, typically investigating, identified, monitoring and resolved (Google’s AI Studio page uses shorter labels such as detected and resolved). Each update has a timestamp.
Google’s AI Studio status page covers the Gemini API and AI Studio. If you use Gemini through Google Cloud, check the Google Cloud page instead.
Using the Gemini app rather than the API? The pages above are for developers. Google’s Workspace Status Dashboard lists Gemini among its services, which helps if you use Gemini through a work or school Google account. We didn’t find a separate public status page for the Gemini app on personal accounts. If you use it that way and the dashboard shows nothing, go to Step 3 and rule out your own side.
Step 2: Check Several Providers at Once
If you use more than one AI service, or your app runs on a cloud platform, it helps to see every status page side by side. Our AI API status page lists 29 AI providers. For most of them (26 on 2026-11-21) it reads the official status page every 15 minutes and shows any open incident and how long recent incidents lasted. For the other three (Google’s Gemini API, Moonshot AI and Z.ai) it links to the official page or notes that none exists, so check Gemini directly (Step 1).
Two points of honesty about that page:
- It shows what the providers post, not our own tests. We don’t send test requests to the APIs, so we never claim to measure uptime. Our methodology page explains this.
- If a provider hasn’t posted an incident yet, we can’t show one. When something is clearly broken but no page says so, the provider may simply not have posted it yet.
Did you know? On 2026-11-21 our page counted 167 incidents posted in the previous 30 days by the 26 providers whose pages it reads. A posted incident does not always mean the service is fully down for you.
Each provider has its own page with its incident history, for example OpenAI and Anthropic.
If your main AI provider went offline for an hour today, what would the people who rely on you see?
Step 3: Rule Out Your Own Side
If the status pages look clean, check these before assuming an outage:
- Try another network or device. A company VPN, a firewall or a browser extension can block one site.
- Sign out and back in to the chat app. Expired sessions cause confusing errors.
- Check your plan’s limits. Chat apps limit how many messages you can send in a period, and APIs have rate limits and spending limits.
- Try a smaller request. A very long document or conversation can hit size limits and fail even when the service is healthy.
Step 4: Read the Error Code (for API Users)
If you’re calling an API, the HTTP status code usually tells you whose problem it is. Both OpenAI’s and Anthropic’s error pages describe these codes:
| Code | What it usually means | Whose side | What to do |
|---|---|---|---|
| 401 | The API key is missing, wrong or revoked | Yours | Check the key and the project it belongs to |
| 429 | Too many requests or tokens in a short time, or a spending or credit limit was reached | Usually yours | Slow down, follow the Retry-After header if one is sent, check your limits and billing |
| 500 | An error on the provider’s servers | Theirs | Retry after a short wait; check the status page |
| 503 | OpenAI: the model is temporarily overloaded | Theirs | Retry with a wait, or switch models |
| 529 | Anthropic: the API is temporarily overloaded | Theirs | Retry with a wait, or switch models |
A sudden wave of 500, 503 or 529 errors across many requests is the clearest sign of a real outage. A steady stream of 429 errors almost always means a limit on your account, not a provider problem.
What to Do During a Real Outage
If you’re a chat user:
- Wait 10–15 minutes and check the status page again. The latest incident update often says whether a fix is in progress.
- Save any long prompt you were writing somewhere else, so you don’t lose it on a refresh.
- If the task is urgent, use another assistant for a while.
If you run an app on an AI API:
- Retry with backoff. Wait a little longer after each failed attempt (for example 1, 2, then 4 seconds) instead of hammering the API.
- Fail gracefully. Show users a clear message, for example the assistant is busy, please try again in a few minutes, rather than a blank screen.
- Have a fallback model. A second model from another provider can carry traffic while the first recovers. Our price board helps you pick one at a similar price.
- Subscribe to updates. Status pages offer email or feed alerts. For example, OpenAI publishes an RSS feed of its incident history at status.openai.com/history.rss, and the Claude status page publishes an Atom feed.
What This Means for You
- Developers: add the providers’ status feeds to your alerts, and treat a burst of 500, 503 or 529 errors as a provider incident: back off and retry rather than hammering the API.
- Founders and team leads: decide in advance what users see during an outage, and whether a second provider is worth having ready.
- Curious beginners: every online service has bad days. Check the status page, wait a little and try again; it usually passes.
Copy-Ready Outage Checklist
- Open the official status page for the product I use
- Find my component (API, chat app, sign-in)
- Check our AI API status page for other providers
- Try another network or device and sign in again
- Read the error code: 429 (my limits) or 500/503/529 (the provider)
- Retry with growing waits, or switch to a fallback model
- Subscribe to the status page’s updates
Bottom Line
- Open the official status page for the product you use, and look at the banner, the components and any open incident.
- If you use several services, scan our AI API status page to see most of them together. Check Gemini on Google’s own pages.
- If nothing is posted, rule out your network, sign-in and limits.
- For API errors, 429 usually points to your account, while 500, 503 and 529 point to the provider.
- During a real outage, wait, retry with backoff, and keep a fallback ready.
For a deeper look at what the labels mean, see our methodology page. Want these updates by email? Sign up here, or follow new guides and data stories through our RSS feed.
