gemini-3.5-flash alias intermittently returns 404 (routes to decommissioned backing build gemini-v4p1s-rev24-ajax-sentinel)

Gemini API
,
1

The published model alias gemini-3.5-flash on the Gemini Developer API (generativelanguage.googleapis.com) intermittently returns 404 NOT_FOUND for a large fraction of generateContent requests. The error names an internal backing build that does not exist:

404 NOT_FOUND

{

“error”: {

"code": 404,

"message": "Model not found: models/gemini-v4p1s-rev24-ajax-sentinel",

"status": "NOT_FOUND"

}

}

The same alias succeeds on the remaining requests, which strongly suggests gemini-3.5-flash is load-balancing across multiple backing builds and one of them (gemini-v4p1s-rev24-ajax-sentinel) is decommissioned/unhealthy and needs to be pulled from rotation.

Impact

Production traffic using gemini-3.5-flash fails roughly two-thirds of the time, intermittently and unpredictably. This is not retryable via normal backoff because it surfaces as a 404, not a 429/503.

Reproduction (minimal, library-independent)

Plain curl against the REST endpoint, minimal request body, no SDK involved:

for i in $(seq 1 25); do

curl -s “https://generativelanguage.googleapis.com/v1beta/models/gemini-3.5-flash:generateContent?key=$GOOGLE_API_KEY” \

-H 'Content-Type: application/json' \\

-d '{"contents":\[{"parts":\[{"text":"hi"}\]}\]}' \\

| grep -o '"message": "\[^"\]\*"' | head -1

done

Observed result: 17 of 25 calls failed (~68%), every failure identical: "Model not found: models/gemini-v4p1s-rev24-ajax-sentinel". The other 8 returned normal completions.

Key facts

  • gemini-3.5-flash is present in GET /v1beta/models and the API key has access to it.

  • A direct generateContent call to gemini-3.5-flash succeeds part of the time, so this is not an access/permissions or model-name issue — it’s specific backing builds behind the alias failing.

  • Sibling/related models are 100% stable in the same test (0 failures out of 12 calls each):

    • gemini-flash-latest

    • gemini-3-flash-preview

    • gemini-2.5-flash

  • Reproduced with the raw REST API (no client library), so it is not a google-genai / langchain-google-genai issue. Also reproduced through the google-genai Python SDK.

Environment

  • API: Gemini Developer API (generativelanguage.googleapis.com), v1beta

  • Auth: API key (AI Studio), not Vertex AI

  • Date/time observed: 2026-05-29, ~22:00 UTC

  • Request body: minimal, no generationConfig / thinkingConfig required to trigger it

Requests

  1. Please pull the unhealthy backing build gemini-v4p1s-rev24-ajax-sentinel out of the gemini-3.5-flash alias rotation.

  2. Confirm whether gemini-3.5-flash is a recommended GA alias or should be avoided in favor of gemini-flash-latest.