Gemini API returns 503 UNAVAILABLE for all requests on one account; works on another account (original) (raw)

February 19, 2026, 11:29am 1

Summary

I’m seeing 503 UNAVAILABLE responses from the Gemini API on every request (100% repro) for one account/project. The same requests succeed immediately when I switch to another Google account (different project / API key). This makes it look account/project scoped rather than a temporary capacity spike.

Error message

503 UNAVAILABLE

{‘error’: {‘code’: 503, ‘message’: ‘This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.’, ‘status’: ‘UNAVAILABLE’}}

What I expected

Occasional 503s during load would be understandable, but not 100% persistent. I expected requests to eventually succeed with retries/backoff.

What actually happens

note: used model is nano banana pro

Mike5 February 19, 2026, 2:26pm 2

Same here, also get the 503 error

koen February 19, 2026, 3:09pm 3

Same here, this has been happening for over a week now

KBoi February 19, 2026, 3:35pm 4

Same here, full of 503 errors today

I’m facing the same issue. It was working fine yesterday, but today I keep getting this error repeatedly.

Berna February 19, 2026, 4:17pm 6

Same problem here. Also get the 503 error on my app

Getting a 503 on all my Gemini

Same, has been especially high for errors over past couple hours for both my G Cloud account and the backup Nano Banana Pro API I use as a fallback.

First starting having issues a few weeks ago, added a fallback option and now both are erroring out. Everyone in the world is using this stuff :exploding_head:.

Even though https://status.cloud.google.com/, it’s lies because in addition to my experience and those in the message board here, even my quite popular backup API provider shows this message on their web interface with Nano Banana Pro:

Screenshot 2026-02-19 at 1.39.23 PM

Nan_Zhang February 19, 2026, 6:45pm 9

Same here for batch API, the current batch job has been stuck in RUNNING for 48+ hours, and logs in AI studio show that every request failed with 503.

florinzf February 20, 2026, 4:22am 10

I have the same problem on Nano Banana Pro. It’s already been quite a few hours since this is happening..

NachoNeko February 20, 2026, 4:22am 11

Same here, also get the 503 error

SAME ERROR

{
“error”: {
“code”: 503,
“message”: “This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.”,
“status”: “UNAVAILABLE”
}
}

downdetecto .com/ status/googlegemini/

Its definitely down

dhruvil February 24, 2026, 2:07pm 14

Same issue from past 15 days i guess and i think it became slow as well. it use to take max 60 seconds for my task previously, but now its taking more than 100 seconds with transient errors. Because of this im hitting cloudflare timeout of 100 seconds. Users are complaining!

*Update: Evidence of False Positive ToS Bans tied to GCP Identity/Session Desync (Even with Native CLI)*

Hi everyone, just wanted to share a critical update on my investigation into the sudden 403 ToS bans impacting Paid/Ultra Tier users.

After migrating completely away from any third-party open-source CLIs and running **exclusively on the native official CLI**, the severe account instability and 403 ToS blocks persisted.

However, while doing a deep audit of my Google Cloud environment, I stumbled upon a major architectural anomaly that I believe is the actual root cause triggering the automated WAF/Anti-Abuse system.

It appears there is a severe “Session Bleed” or identity desynchronization occurring in the backend when a user operates a standard Consumer Account (e.g., standard Google account) alongside a Workspace/Cloud Identity organizational account in the same environment.

When the official native CLI attempts the OAuth handshake, the Application Default Credentials (ADC) or the active session token gets polluted with an inconsistent mix of organizational constraints and consumer tier limits.

**The Result:** The automated security backend reads this “schizophrenic” token state not as a legitimate Google Cloud UI/CLI bug, but erroneously flags it as a “Violative/Unauthorized Client” or an attempt to bypass ToS, triggering an immediate and silent 403 ban on the Gemini API.

This means many of us aren’t being banned for “abusing” the API or using “unauthorized tools.” We are victims of a False Positive generated by a session management bug within GCP’s own identity infrastructure crossing wires between our Consumer and Workspace profiles.

I am currently compiling a formal architecture vulnerability report for the Google Cloud VRP (Vulnerability Reward Program) regarding the underlying GCP Identity flaw. But the AI team needs to urgently halt these automated ToS bans on Paid Tier accounts until the Cloud Infrastructure team resolves this cross-account authorization bleed.

The collateral damage on legitimate developers is entirely unjustified.

-–

*(Note: I am holding back the exact reproduction steps of the GCP bug as it is being routed to the Cloud VRP, but this context is vital for the AI team to understand the false positives).*

Same issue here, after havig no problems using my API key in a project on my personal google account, after moving the project to a new Workspace account, immediate issues. Paid account, no difference. This is really hard to swallow when trying to develop commercial application…

503 UNAVAILABLE. {‘error’: {‘code’: 503, ‘message’: 'This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later.

Hey

503 (“Service Unavailable”) errors are unrelated to your quota and indicate our services are temporarily overloaded and can’t service your request. These errors are more common during certain times of the day. We apologies for the impact of these errors & are working hard to provision more resources and reduce their occurrence.

i changed the Gemini API key back to one created on my personal (paid) account and now i do not have any 503s. On my business workspace account it’s like 30% change of hitting a 503. Is there anything I can change to this business account (since i want the cost billed there..)

Stokes March 1, 2026, 9:35pm 20

In case this helps anyone else…. I was getting this 503 for all calls, which just started failing suddenly during the middle my trial. When I went into my account settings, I noticed that Vertex AI wasn’t enabled.

I know it was enabled for this account, because I’ve been using the API for weeks. But I just re-enabled it and the 503s went away.