# Gemini status

> Is Gemini down? Live status for the Gemini API, uptime history, and per-region response times from our own probes every 2 minutes.

Published: 2026-09-08 | Canonical: https://yoping.me/status/gemini

## How we check Gemini

We watch generativelanguage.googleapis.com/v1/models from all five of our probe regions, every 2 minutes. A region that sees a failure does not make this page say "down" on its own: a second region has to agree first, the same confirmation rule every YoPingMe monitor uses.

We watch the API host applications call to run Gemini models, and pin the response an unauthenticated request receives rather than assuming a 2xx. The consumer chat app at gemini.google.com is served separately, and the API is what an outage actually breaks for a product built on Gemini.

The live board - the current verdict, 24-hour, 7-day, and 30-day uptime, and per-region response times - is on the HTML page at https://yoping.me/status/gemini. Those numbers change too fast to repeat honestly in a static mirror, so where an answer below says "the board above", it means that page.

## What is Gemini?

Gemini is Google's family of AI models, offered both as a consumer chat
product and as an API that applications call directly to generate text,
analyse images, or build features on top of. Those two surfaces are
served separately, and this page watches the API specifically, because
that is the dependency an application actually has, distinct from
whether the consumer chat interface happens to be working at any given
moment.

Rate limiting is the most common cause of Gemini API errors that get
mistaken for an outage. Limits are enforced per model and per project,
across both request counts and token volume, and a traffic spike or a
batch job can exhaust the allowance while the service answers every
other caller normally. The response identifies which limit was hit,
which is the fastest way to tell throttling from an actual problem.

Latency is worth tracking separately from availability, and it is why
the per-region response times above sit next to the up or down reading
rather than behind it. A request that succeeds but takes noticeably
longer than usual is a real degradation for anything built around fast
responses, even though a monitor that only checked whether requests
succeeded would report a perfect day through exactly that kind of
incident.

Model availability is the second thing worth separating from platform
health. Google retires older model versions on a published schedule, so
a request naming a model that has been deprecated returns an error while
every current model keeps working normally. The error names the specific
model, and integrations that log only a generic failure hide the detail
that would otherwise turn a one-line configuration fix into a longer
investigation.

The practical question for anything built on the Gemini API is what
happens when a request fails or takes too long. A feature that calls a
model directly in the request path has made that model a hard dependency
of the response, and a fallback, a cached answer, or an honest error
message beats a request that hangs indefinitely. Deciding that in
advance is far easier than working it out during the first real
incident.

## Frequently asked questions

### Is Gemini down right now?

Check the board above; it reflects our own probes against the Gemini API from all five of our regions, refreshed every couple of minutes and confirmed across two regions before this page would call it down.

### I am getting 429 errors from the Gemini API. Is that an outage?

No, a 429 means the request reached the API and was throttled, not rejected as broken. Limits apply per model and per project, and a burst of traffic or a batch job can exhaust them while the service is entirely healthy. The response identifies which limit was hit, which settles whether you are being throttled or something is actually wrong.

### My responses are slower than usual but every request still succeeds. Should I treat that as an incident?

Not as downtime, but it is worth tracking separately, which is why the response times above sit alongside the up or down verdict. Latency varies with model, prompt length, and current demand, and a degradation that never fails a request would look like a perfect day to a monitor that only checked availability.

### Does a model being deprecated mean Gemini is down?

No. Google periodically retires older model versions on a published schedule, and a request naming a retired model returns an error while every current model responds normally. The error names the model, and Google's model documentation lists what is currently available.

### Does the Google Workspace Status Dashboard cover the Gemini API too?

Not specifically. That dashboard covers Workspace products like Gmail and Drive, and it reports what Google has confirmed internally. This page reports what our own probes measured against the Gemini API from outside Google's infrastructure, on a fixed schedule, whether or not an update has been posted anywhere.