Provider status · Google (Gemini)
Is the Gemini API down right now?
As of 26 Oct 2026, 11:20 UTC · one real completion to generativelanguage.googleapis.com every 10 minutes, on our own account
This is the Gemini API as an application sees it: generativelanguage.googleapis.com, not the Gemini app, Vertex AI or the Gemini exchange. Every check we could make is counted; every check we could not make is listed as not measured and counts as neither uptime nor downtime.
Free account · provider alerts on any plan · an email after 20 minutes without an answer, and again when it comes back
Skips a provider that is not answering and routes to the next best model · your own keys · 1,000 requests a month free
- 100%
- of the 142 checks we could make in the last 24 hours were answered
- 100%
- answered over the last 7 days — 1,010 measured of 1,010 checks
- 590 ms
- median reply to one short completion over the last 24 hours, from our machine in Germany
Last 24 hours and 7 days
Gemini API uptime over the last 24 hours and 7 days
“Answered” counts only the checks we could make, and the denominator is printed beside every figure. The median reply is our own round trip to generativelanguage.googleapis.com.
- Last 24 hours
- 100% of checks answered — 142 measured of 142 checks.
- Last 7 days
- 100% of checks answered — 1,010 measured of 1,010 checks.
- Median response, last 24 hours
- 590 ms to one short completion on our own key, from one machine in Germany. The median, not the mean, so one slow check from a machine that is often running a benchmark sweep at the same time does not describe the API.
- Last check
- 26 Oct 2026, 11:20 UTC: answering in 711 ms.
Hour by hour, last 7 days
One cell per hour, in UTC. A check that failed on our side is hatched and counts as neither uptime nor downtime; an hour with no check stays faint, so a gap in the record shows as a gap.
The record
Gemini API outages in the last 7 days
An outage here is a run of checks the Gemini API did not answer, with the provider’s own words for why. Only provider-caused failures are listed; anything that failed on our side is not an outage.
None in the last 7 days: every check we could make was answered.
Method
What does this page measure, and what does it not?
One real completion, not a ping and not a mirror of Google’s own status page. The same check goes to every provider we benchmark, so every provider’s page is comparable.
- The check
- One short completion on our own key, from one machine in Germany. It asks for a short reply and runs every 10 minutes; a provider that has not answered within 60 seconds is recorded as not answering. The same call goes to every provider, so the pages are comparable.
- The endpoint
generativelanguage.googleapis.com— the Gemini API via Google AI Studio (generativelanguage.googleapis.com), not the Gemini app, not Vertex AI, and not the Gemini exchange, which has nothing to do with Google.- Not probed
- We do not probe the Gemini app, Vertex AI or the Gemini exchange. This page says nothing about those surfaces. For them, see Google’s own status page, aistudio.google.com/status.
- When a check fails
- We ask again fifteen seconds later and record the second answer, so one dropped connection is not published as ten minutes of outage. If the provider still cannot be reached, we test our own side before blaming theirs: whether our server stalled while the check was in flight, whether other providers failed to connect at the same moment, and whether two independent control endpoints answer. Any of those means the failure was ours: it is published as not measured and counts as neither uptime nor downtime.
- What “answered” means
- The provider returned usable content. It does not mean the model was correct, fast or good; that is what the benchmarks are for.
- Where from
- One probe, every 10 minutes, from one machine in Germany, on one account. A provider can be down for us and up for you, or the reverse; if your own traffic disagrees with this page, trust your own traffic.
- The same data by API
- GET
/api/v1/providersand GET/api/v1/analytics/provider-reliabilityreturn these figures with a free Data API key; both are on the API reference.
Reference
What do Gemini API error codes mean?
What the codes mean in the providers’ own documentation, so an account problem is not mistaken for an outage. A reference, not a measurement from our probe.
| Code | Meaning | What to do |
|---|---|---|
401 | Authentication failed | The API key is missing, wrong or revoked. The service is up; the request is not allowed in. Check the key and the account it belongs to. |
403 | Not permitted | The key is valid but may not use this model, endpoint, region or organisation. An account setting, not an outage. |
404 | Unknown model or endpoint | The model name is misspelt or has been retired, or the path is wrong. Check the provider’s current model list. |
429 | Rate limit or quota | Too many requests or tokens in the window, or a spending limit reached. A rate limit passes in seconds: back off and retry. A quota or billing message means the account, not the provider. |
500 | Provider error | Something failed inside the provider while handling the request. Retry with exponential backoff; if it persists across requests, the provider has a problem. |
502 · 503 · 504 | Unavailable, overloaded or timed out | The provider or its edge is overloaded, restarting or too slow to answer. “High demand” and “overloaded” messages are capacity signals: retry with backoff, or route the request to another provider. |
On the Gemini API a 503 that says the model is overloaded or in high demand is capacity on Google’s side, while a 429 is quota on your own project: one is the provider, the other is your account.
During an outage
What should I do when the Gemini API is down?
Confirm it is the service and not your account, retry the way the codes above say, and route around it until it answers again.
01
Separate the account from the service
A 401, 403 or a 429 that mentions quota or billing is your key or your account, and no status page will show it. A 5xx, a timeout or a 429 rate limit is the service. This page shows what generativelanguage.googleapis.com answered to our own account at the same time.
02
Retry with backoff, briefly
For 429 and 5xx, wait and retry with exponential backoff and a cap. If the failures persist across requests, stop retrying: you are adding load to a provider that is already struggling and latency to your own users.
03
Route around it
The Smart Router is one OpenAI-compatible endpoint over every provider we benchmark. It skips a provider that is not answering and sends the request to the next best model, on your own provider keys, with your fallback order if you set one. 1,000 routed requests a month are free.
04
Be told when it is over
Turn on provider outage emails: one when a provider you track or have connected keys for stops answering our checks for 20 minutes or more, and another when it answers again. Free account, any plan.
Skips a provider that is not answering and routes to the next best model · your own keys · 1,000 requests a month free
Free account · provider alerts on any plan · an email after 20 minutes without an answer, and again when it comes back
Models
Which Gemini models does AI Stupid Level benchmark?
Every Google (Gemini) model on the board is called through generativelanguage.googleapis.com and gets the same tasks as every other vendor’s models: coding every four hours, reasoning and tool use daily, with an hourly canary probe. Each has its own page with its 24-hour scores and places.
All Gemini models on the same testsIs Gemini getting worse?Compare models head to head
FAQ
Questions about Gemini API status
Answers computed from the feed as of 26 Oct 2026, 11:20 UTC.
Is the Gemini API down right now?
As of 26 Oct 2026, 11:20 UTC: Answering — checked 6 minutes ago, median 590 ms over the last 24 hours. 100% of the 142 checks we could make in the last 24 hours were answered, and 100% of 1,010 over the last 7 days. We check every 10 minutes and this page is rebuilt from the feed about once a minute.
How do you check whether the Gemini API is down?
One short completion on our own key, from one machine in Germany. Every 10 minutes we send one short request to generativelanguage.googleapis.com and record whether usable content came back and how long it took; a provider that has not answered within 60 seconds is recorded as not answering. A failed check is tried again fifteen seconds later and only the second answer is recorded, so a single dropped connection is never published as ten minutes of outage. "Slow" means the reply took longer than 5 seconds from our machine, so the median sits next to it. The record is kept for 7 days.
Why does it say "not measured"?
Because the check failed on our side, not the provider's: an issue with our own account, our own request or our own server (its event loop frozen while the check was in flight, several providers failing to connect at the same moment, or two independent control endpoints not answering either). We publish those checks as not measured rather than hiding them, and they count as neither uptime nor downtime, so a problem of ours never shows up as Google's downtime and the uptime figure only ever divides real observations.
Is this the same as aistudio.google.com/status?
No. aistudio.google.com/status is Google's own report of its own services, incident by incident. This page is an independent measurement: one real completion to generativelanguage.googleapis.com every 10 minutes from our own account, with every check we could make counted and every check we could not make listed separately. The two can disagree; when they do, the one that matches your own traffic is the one to trust. For the Gemini app, Vertex AI or the Gemini exchange, which we do not probe, aistudio.google.com/status is the place to look.
Does this cover the Gemini app, Vertex AI or the Gemini exchange?
No. We probe generativelanguage.googleapis.com only: the Gemini API via Google AI Studio (generativelanguage.googleapis.com), not the Gemini app, not Vertex AI, and not the Gemini exchange, which has nothing to do with Google. For the Gemini app, Vertex AI or the Gemini exchange, see aistudio.google.com/status.
Can I get an alert when the Gemini API goes down?
Yes, on any plan. A free account can turn on provider outage emails in account settings: when a provider you track or have connected keys for stops answering our checks for 20 minutes or more you get an email, and another when it answers again. The Smart Router handles the outage itself: it skips a provider that is not answering and routes the request to the next best model, on your own provider keys, with 1,000 routed requests a month free.
Every provider
Other providers we check every 10 minutes
Hear it from the probe, not from your users
Free account: every ranking over 24 hours, 7 days and 30 days, a 3-model watchlist with email alerts, 1,000 routed requests a month and a Data API key. No card.
Provider outage emails on any plan, for the providers you track or have connected keys for
Free account · 24 hours, 7 days and 30 days · no card
Already have an account? Log in