OpenAI models on Google Cloud are now available. If your company runs on Google Cloud, you may no longer need a second vendor contract to use GPT. OpenAI is adding its frontier models to Vertex AI Model Garden, alongside Gemini, Claude, Llama and others, so you get the same IAM controls, the same invoice and the same deployment tooling. Most teams that standardized on Google Cloud have been running two AI stacks: Gemini on Vertex and a separate OpenAI account elsewhere. The launch is in private preview today, and general availability is promised “in the coming weeks.” Start planning now.
What’s actually new with OpenAI models on Google Cloud
On October 21, 2026, OpenAI announced it would bring its models to Google Cloud, which CNBC called a further expansion beyond Microsoft. The first lineup is GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. Outlets spell the names slightly differently.
The model lineup
- Astra is the heavy model for coding, computer use and science work.
- Sol is pitched as close to Astra in capability at a much lower price.
- Luna is the fast, cheap option for high-volume everyday tasks.
OpenAI also said Codex is coming to Google Cloud on a similar timeline.
Billing and commitments
The commercial details matter more than the model names. According to Techzine, you buy access through Google Cloud Marketplace, usage appears on your normal Google Cloud invoice, and spend draws down your existing Google Cloud commitments. For enterprises with large committed-use agreements, that changes the math: GPT usage can now burn down money already promised to Google instead of becoming a new line item for procurement to approve.
Availability
Access is gated for now. The models are in private preview for selected enterprise customers, with general availability expected within weeks. This launch puts OpenAI models on all three major hyperscalers. Amazon started hosting OpenAI models in April, and Microsoft remains OpenAI’s primary cloud partner. Until now, OpenAI’s only presence in Model Garden was its open-weight gpt-oss models. This is the first time its proprietary frontier models are on Google’s platform.
Why it matters
- One bill, one contract. GPT on Vertex AI runs through Marketplace and counts toward your Google Cloud commit. You skip the separate OpenAI enterprise agreement, security review and invoice.
- The OpenAI–Microsoft exclusivity era is over. OpenAI is now a multi-cloud model vendor. Expect more competition on price, regions and enterprise features.
- Real multi-model AI on Google Cloud. You can route between GPT and Gemini in the same project, with the same service accounts and the same VPC Service Controls perimeter. Per-task model selection becomes an engineering decision, not a procurement fight.
- Less vendor lock-in. If your prompts and evals run on both model families, you have leverage. When a model regresses or a price changes, you swap a model ID instead of migrating platforms.
- Simpler data governance. Security teams that already approved Vertex AI for Gemini have a much shorter path to approving GPT on the same platform. Confirm the data-residency and data-usage terms for partner models before assuming they match Gemini’s.
- Pressure on Gemini pricing. Google now hosts a direct competitor to its own flagship model. Expect sharper pricing and faster releases on both sides.
How to use OpenAI models on Google Cloud today
These steps follow how Vertex AI already serves partner models through its managed “model as a service” setup. Exact model IDs and regions for the GPT-6 family will appear on each model’s Model Garden card. Copy them from there; don’t guess.
1. Request preview access
In the Google Cloud console, open Vertex AI, then Model Garden, and search for OpenAI. If your organization isn’t in the private preview yet, use the request-access option on the model card and contact your Google Cloud account team. Enterprises with existing commits get priority.
2. Enable the APIs and authenticate
Turn on the Vertex AI API in the project you’ll bill against.
gcloud config set project YOUR_PROJECT_ID
gcloud services enable aiplatform.googleapis.com
gcloud auth application-default login3. Enable the model through Marketplace
On the model card, click Enable. This accepts the Marketplace terms and connects billing to your Cloud account. Only someone with billing permissions can complete this step, so line them up ahead of time.
4. Grant least-privilege access
Give the calling service account the Vertex AI User role and nothing broader.
gcloud projects add-iam-policy-binding YOUR_PROJECT_ID \
--member="serviceAccount:llm-app@YOUR_PROJECT_ID.iam.gserviceaccount.com" \
--role="roles/aiplatform.user"5. Make a first call with curl
Vertex AI exposes an OpenAI-compatible Chat Completions endpoint, so the request shape will look familiar. Swap in the model ID and region from the model card.
export PROJECT_ID=YOUR_PROJECT_ID
export LOCATION=us-central1
export MODEL_ID="openai/MODEL_ID_FROM_MODEL_CARD"
curl -X POST \
-H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \
"https://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/endpoints/openapi/chat/completions" \
-d '{
"model": "'"${MODEL_ID}"'",
"messages": [
{"role": "system", "content": "You are a concise senior engineer."},
{"role": "user", "content": "Explain the tradeoffs of Pub/Sub vs Cloud Tasks in 5 bullets."}
],
"max_tokens": 600
}'6. Point the OpenAI Python SDK at Vertex
If your code already uses the openai library, you mostly change the base URL and the credential. Install the dependencies:
pip install openai google-authThen configure the client:
import google.auth
import google.auth.transport.requests
from openai import OpenAI
PROJECT_ID = "YOUR_PROJECT_ID"
LOCATION = "us-central1"
MODEL_ID = "openai/MODEL_ID_FROM_MODEL_CARD"
creds, _ = google.auth.default(scopes=["https://www.googleapis.com/auth/cloud-platform"])
creds.refresh(google.auth.transport.requests.Request())
client = OpenAI(
base_url=f"https://{LOCATION}-aiplatform.googleapis.com/v1/projects/{PROJECT_ID}/locations/{LOCATION}/endpoints/openapi",
api_key=creds.token, # short-lived token; refresh it in long-running services
)
resp = client.chat.completions.create(
model=MODEL_ID,
messages=[{"role": "user", "content": "Summarize this incident report in 3 bullets: ..."}],
)
print(resp.choices[0].message.content)7. Check quotas before you launch
Partner models on Vertex have their own per-model, per-region quotas, separate from Gemini’s. Open IAM and Admin, then Quotas, filter on the model name, and file an increase request early. Preview quotas are usually low.
8. Run a head-to-head eval before choosing
Run the same prompt set against GPT and Gemini. A simple routing config keeps the decision reversible:
# model_routing.yaml
routes:
code_review: openai/GPT_ASTRA_ID # heavy reasoning on code
support_replies: openai/GPT_LUNA_ID # high volume, low cost
long_doc_qa: google/gemini-2.5-pro # very long context
default: google/gemini-2.5-flashHow it compares: GPT vs Gemini on Vertex AI and beyond
The launch-day pricing below comes from one early comparison table. Treat it as indicative and check the live Model Garden pricing page before you budget.
| Option | Models | Reported price (input / output per 1M tokens) | Billing | Best fit |
|---|---|---|---|---|
| GPT-6 Astra on Vertex AI | OpenAI flagship | $10.00 / $50.00 | Google Cloud Marketplace, counts toward commit | Hard coding, agents, science work |
| GPT-6.1 Sol on Vertex AI | OpenAI mid-tier | $2.00 / $10.00 | Google Cloud Marketplace | Near-flagship quality at about a fifth of the cost |
| GPT-6 Luna on Vertex AI | OpenAI fast tier | $0.10 / $0.50 | Google Cloud Marketplace | Classification, extraction, chat at scale |
| Gemini on Vertex AI | Google first-party | See Vertex pricing | Native Vertex billing | Long context, multimodal, Google Workspace and Search grounding |
| Azure OpenAI / Microsoft Foundry | OpenAI models | See Azure pricing | Azure commit | Microsoft-centric shops; usually gets new OpenAI releases first |
| OpenAI API direct | Full OpenAI catalog | See OpenAI pricing | Separate OpenAI account | Startups, newest features, Assistants and Responses tooling |
Our view: the biggest gain isn’t any single model’s quality. It’s picking a model per task without leaving your cloud. Gemini still wins on native integration: grounding, very long context, multimodal input and Workspace. GPT is the stronger option for teams whose prompts, evals and agent code were built around OpenAI’s behavior. Those teams can now bring that work into Google Cloud without rewriting it.
What’s next
General availability
Watch the move from private preview to GA and what ships with it. GA should bring wider regional availability, published quotas and, most importantly, clear terms on data residency, provisioned throughput and SLAs. If provisioned throughput launches for the GPT models at GA, latency-sensitive production workloads become realistic. Until then, use preview access for prototyping and evals.
Feature parity
Model access is one thing; the full OpenAI developer surface is another. Structured outputs, tool calling, the Responses API, realtime voice and fine-tuning may not all arrive at once. Partner models on cloud marketplaces often trail the vendor’s own API by weeks or months. Codex matters here too: if it plugs into Cloud Build, Cloud Workstations or Gemini Code Assist workflows, it becomes a serious option for engineering teams standardized on Google tooling.
The multi-cloud race
OpenAI now sells through AWS, Microsoft and Google, and all three are competing to be the default place enterprises run every frontier model. Expect faster model drops in Model Garden, more aggressive commit-based discounts, and Google pushing Vertex’s evaluation and routing tools to keep multi-model customers on its platform.
Frequently Asked Questions
Can I use OpenAI models on Google Cloud right now?
Only if you’re in the private preview, which is limited to selected enterprise customers at launch. OpenAI says general availability is coming in the next few weeks. Ask your Google Cloud account team for access now so your project is ready at GA.
How is GPT on Vertex AI billed?
Through Google Cloud Marketplace. Usage appears on your normal Google Cloud invoice and counts toward existing Google Cloud spend commitments. You don’t need a separate OpenAI account or invoice.
Is my code compatible if it already uses the OpenAI SDK?
Mostly. Vertex AI exposes an OpenAI-compatible Chat Completions endpoint, so you change the base URL, swap the OpenAI API key for a Google Cloud access token, and use the Vertex model ID. Test any advanced features you rely on, because not every OpenAI-only API is guaranteed on Vertex at launch.
Should I pick GPT or Gemini on Vertex AI?
Run your own evals; the shared platform makes them cheap. As a rough rule, use Gemini for very long documents, multimodal input and Google ecosystem grounding. Use GPT where your prompts and agents were already tuned on OpenAI models or where it scores better on your coding and reasoning benchmarks. Route by task instead of picking one winner.
Does this end OpenAI’s relationship with Microsoft?
No. Microsoft remains OpenAI’s primary cloud partner. What changed is exclusivity: OpenAI models now run on AWS, Azure and Google Cloud, making OpenAI a multi-cloud model vendor rather than an Azure-only one.
Are quotas shared with my Gemini quotas?
No. Partner models on Vertex AI have their own per-model, per-region quotas. Check them on the Quotas page in the console and request increases early, because preview limits are usually conservative.
Go deeper than this article
This article covers the essentials. Our Technical & Coding eguide collection gives you the full step-by-step playbooks — prompts, workflows, and copy-paste recipes built for exactly this work.
