Gemini 3.8 Flash API returns 503, but AI Studio works

My free-tier Gemini API project repeatedly receives HTTP 503 UNAVAILABLE when calling gemini-3.8-flash through generateContent.

The error says: “This model is currently experiencing high demand.”

A minimal one-line API request also fails, but Gemini 3.8 Flash successfully responds in Google AI Studio. The status page reports all systems operational.

Could you help determine whether this is model capacity, project access, or a configuration issue?

I am also experiencing the same issue, so for now for local development, I am currently using gemini-3.5-flash-lite model .

For the following content I am getting Repeated 503 UNAVAILABLE error on free tier

{
  "title": "Understanding Redis Caching",
  "content": "Redis is an in-memory data store commonly used for caching frequently accessed data. It can improve application performance by reducing repeated database queries.",
  "instruction": "Make the suggestions suitable for backend developers."
}

after using gemini-3.5-flash-lite for free tier, solving the current development implementation.

for the same request I am getting this response .

{
    "status": "success",
    "data": {
        "suggestedTitle": "Mastering Redis Caching for Backend Performance",
        "summary": "An introduction to Redis as an in-memory data store, focusing on how backend developers can leverage caching to minimize database load and accelerate application response times.",
        "topics": [
            "Redis",
            "Caching",
            "Backend Development",
            "Database Optimization",
            "Performance Tuning"
        ]
    }
}

Hi, 503 errors are due to our services being temporarily overloaded. Please see this pinned post for suggestions on how to handle this.

I can reproduce this issue with gemini-3.8-flash on a free-tier project.

Gemini 3.8 Flash responds in approximately 2–3 seconds in the AI Studio web interface. However, minimal API requests fail. Gemini 3.5 Flash Lite API requests succeed with the same API key and network.

I reproduced the problem outside my application, using direct REST requests through HTTPX without the Gemini SDK.

Endpoint: POST https://generativelanguage.googleapis.com/v1beta/interactions
Input: “Reply with OK.”
Configuration: max_output_tokens=65536, thinking_level=“medium”, store=false.
Each diagnostic sends one request, with no retries and a 45-second local deadline.

Results on October 22, 2026:

  • Non-streaming: HTTP 503 with a high-demand error after 5.093 seconds. TCP/TLS and request transmission completed within 0.562 seconds.

  • Streaming, started at 21:25:27.891 UTC: HTTP 200 headers arrived after 15.656 seconds, but no complete JSON SSE event or usable answer was processed before the 45-second local deadline.

No HTTP_PROXY, HTTPS_PROXY or ALL_PROXY environment variables were present.

Could the team investigate whether this involves backend capacity, project-specific access or routing? No request ID was captured. Please advise which additional identifiers you need and how to share project details securely.

I’m experiencing the same issue with the Gemini API. For the past 4–5 days, my application has had intermittent failures with 503 Service Unavailable errors. Sometimes requests work normally, but other times they fail repeatedly and only succeed after retrying. The issue is inconsistent and is affecting my production app. Is anyone else experiencing this with gemini-3.1-flash-lite?

Today I duplicated a project using AI Studio’s own tool. Copied code/repo Gemini API calls get permanent 503 errors, while original code works like a charm → gemini-3.8-flash

Did not test with other models, will downgrade until fixed if they work.

That has happened to me too.