Enterprises with committed Google Cloud spending are getting another way to pay for OpenAI models: through Google Cloud Marketplace, using budgets they have already committed to Google.
In Google Cloud’s October 21 announcement, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are entering Model Garden in private preview for selected enterprise customers. The preview also supports connecting local Codex CLI and Desktop workflows to Google’s platform for inference.
Access is restricted. Integrations with Google Antigravity and Gemini for business are still coming, and some advertised security capabilities will not be available at the outset.
Even with those limits, the competitive shift is substantial. Google is becoming a purchasing and deployment channel for models that compete with its own Gemini offerings. OpenAI gains another enterprise distribution route while, according to OpenAI, Microsoft remains its primary cloud partner.
Private Preview Comes Before the Broader Rollout
Selected enterprise customers can begin using the three OpenAI models through Google’s APIs and connect their local Codex workflows to its platform. Other interested customers must register their interest; access is not unrestricted.
The rollout has several distinct stages:
- Starting now: Private-preview access to GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna for selected enterprises.
- Supported during the preview: API access and connections from local Codex CLI and Desktop workflows to Google’s platform for inference.
- Still ahead: OpenAI model availability inside Google Antigravity and Gemini for business.
- Not yet generally available: The broader offering, which OpenAI says is planned for the coming weeks without specifying a launch date.
Google calls the forthcoming business integration “Gemini for business,” while OpenAI refers to the Gemini Agent. Neither reference establishes that the integration is available today.
The Codex support also has limits. Connecting a local CLI or Desktop workflow to Google’s inference service gives developers a way to use those interfaces with a cloud-backed model endpoint. The model does not run locally. That support also does not establish that every Codex feature is included or that the forthcoming Antigravity integration is operational.
Buyers first need to confirm whether their organization can enter the preview and whether it supports their required workflow. A planned general release offers no guarantee of current access.
Existing Google Cloud Commitments Change the Buying Decision
The clearest commercial benefit is the ability to purchase OpenAI access through an existing cloud relationship.
Google says transactions will run through Google Cloud Marketplace, usage will appear on a single Google Cloud invoice, and spending can draw down existing cloud commitments. Usage also counts toward Flexible Savings Plans across model providers.
For an enterprise already committed to spending with Google Cloud, this can change deployment economics before model performance enters the discussion. Buying OpenAI access separately may require an additional budget allocation while leaving a Google commitment untouched. Through Marketplace, eligible model usage can contribute toward that commitment.
The procurement advantage does not establish a lower model price. Google’s announcement discloses no token rates, detailed discount terms or per-model regional availability list, so customers cannot use it alone to calculate whether this route costs less than another purchasing channel.
Finance and procurement teams will need to check how their agreement treats the purchase, including applicable model rates, how usage is credited against commitments and how Flexible Savings Plans affect the final bill.
A consolidated invoice can be valuable even without the lowest unit price. Some organizations may prefer the Google route because it fits their existing budget and purchasing process; others may find that their workload, contract terms or access requirements favor a different channel.
Three Model Options, but No Performance Comparison
Google presents the initial model lineup as a set of choices for different workloads:
- GPT-6 Astra: Intended for demanding coding, computer-use, professional and scientific research tasks.
- GPT-6.1 Sol: Positioned as offering near-Astra intelligence at a lower cost for agentic coding, computer use and professional work.
- GPT-6 Luna: Positioned as a fast, cost-efficient option for everyday work at scale.
These descriptions come from the vendor. The announcement provides no independent performance findings, benchmark results, latency measurements or prices with which customers could verify the proposed tradeoffs.
“Near-Astra intelligence,” in particular, does not establish how Sol performs on a company’s own coding repository, document collection or agent workflow. Calling Luna cost-efficient reveals nothing about its cost per successfully completed task.
Enterprises evaluating the preview should compare the models on representative work, including failure rates and the amount of human correction required. A lower-priced response can become expensive if an agent repeatedly retries a task or produces output that needs extensive review.
Google argues that its platform lets customers select and swap models according to performance, latency and cost. Centralizing access can make that process easier to organize, though prompts, tool behavior and application assumptions still need validation whenever the underlying model changes. The models are not interchangeable.
Governance Controls Have Important Launch Boundaries
Security is a major part of Google’s pitch. The company says customer prompts, context and outputs remain within the designated data perimeter and are not used to train foundation models.
OpenAI models also run on dedicated Google Cloud infrastructure, according to Google, with queries routed through isolated gateways. The announced controls include authenticated identities, VPC Service Controls protection and unified audit logging.
For enterprises, adding a model provider can raise questions about separate identity, auditing and data-control arrangements. Google presents the OpenAI offering as part of its existing governance environment.
Two capabilities are explicitly described as coming soon in Google’s governance description: Zero Data Retention options and Confidential Compute support. Buyers should exclude both from their assessment of available launch features.
Sources
- Google Cloud’s October 21 announcementcloud.google.com
