Allowlist request: gemini-3-pro-image-preview & gemini-3.1-pro-preview on Vertex AI — existing AI Studio customer ready to migrate

Gemini API
1 
Hi Google Cloud team,

Requesting allowlist access for the following Vertex AI preview models on our production projects:

- `gemini-3-pro-image-preview` — image generation

- `gemini-3.1-pro-preview` — multimodal vision analysis

## Customer context (AI Studio)

We're an active Gemini API customer on **Google AI Studio**, with billing enabled and ~4,000 production image generations/month sustained on `gemini-3-pro-image-preview`, plus the corresponding multimodal analysis calls on `gemini-3.1-pro-preview`. Happy to share our billing account ID and historical usage metrics privately to verify standing.

We're looking to migrate this workload from AI Studio to Vertex AI to take advantage of GCP-native billing, IAM-based access control, and Cloud Audit Logs.

## Project details (Vertex side)

- **GCP Project IDs:** two production projects under our org, shared privately upon staff response

- **Organization:** yeetystudios.com

- **Endpoint:** `global`

- **Billing:** enabled, paid account

- **Vertex AI / Agent Platform API:** enabled on both projects

- **Service account role:** `roles/aiplatform.user` (Vertex AI User) provisioned on both projects

- **GA model verification:** `gemini-2.5-flash-image` confirmed working end-to-end on Vertex

## Workload shape

Production multimodal image-generation pipeline. Each `generateContent` call is multi-reference, typically carrying **7–14 reference images** via `inline_data` parts (we hit the documented per-request input cap), with `imageSize: "2K"` output. Vision analysis calls run alongside on `gemini-3.1-pro-preview`.

- **Current volume:** ~4,000 image generations/month sustained on AI Studio

- **Target:** 5–10× over the next quarter

## Errors we see on Vertex preview models without allowlist

Routing the same body to `gemini-3-pro-image-preview` on Vertex produces unstable behaviour consistent with capacity-rationed pre-allowlist access:

1. **Bare HTTP 404 `NOT_FOUND`**, empty `details[]`:

```json

{"error":{"code":404,"message":"Requested entity was not found.","status":"NOT_FOUND"}}

  1. HTTP 429 RESOURCE_EXHAUSTED despite the Quotas dashboard reporting the relevant per-project metric as Unlimited / Adjustable: No.

  2. Intermittent 200 / 429 / 404 on byte-identical bodies sent sequentially — capacity tracks rather than request shape.

  3. gemini-2.5-flash-image on the same SA returns 200 reliably, confirming credentials, IAM, project setup, and request shape are correct.

This blocks us from completing our migration from AI Studio.

Request

Please grant our two production projects (IDs to follow privately) full allowlist access to gemini-3-pro-image-preview and gemini-3.1-pro-preview — i.e. first-class quota, not the rationed-overflow capacity we appear to be receiving today. Our current AI Studio volume is ~4,000 generations/month and we expect to scale 5–10× over the next quarter, so unrationed access matters for production stability.

Happy to provide privately:

  • AI Studio billing account ID and historical usage statistics

  • Sample (PII-free) request/response logs

  • Security review documentation

Thanks for your time.

2

Hello @yeety_adhd,
This forum is primarily intended for the Gemini API and Google AI Studio. For issues related to the Vertex AI API, can you contact your sales representative if you have one? Otherwise, you may find more relevant assistance on theGoogle Cloud Forum.