Claude and OpenAI on Google Cloud: Pricing, Billing and Commitments

October 23, 2026. Anthropic now sells its full Claude API through Google Cloud at the same list prices as its own, from $0.10 per million input tokens on Claude Haiku 5.5 to $10 on Claude Fable 5.1, with the charges on your Google Cloud invoice and counted toward your Google Cloud spend commitment. Anthropic made its new Anthropic-managed option generally available on October 22, a day after OpenAI said its models are coming to Google Cloud's Model Garden, where Google says GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out in private preview to select enterprise customers. For a business that already pays Google Cloud, Claude and, once OpenAI's rollout opens up, GPT models can go through one bill and one set of access controls.

What Anthropic, OpenAI and Google announced

1. Claude Models on Google Cloud (Anthropic managed), generally available October 22. It is the Claude API, run by Anthropic and reached through your Google Cloud project, per Anthropic's announcement. Access goes through Google Cloud IAM, billing goes through Google Cloud Marketplace, and new Claude Platform features and betas ship the same day they reach Anthropic's own API. The announcement lists US and Global endpoints and four current models: Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5.
2. The Google-managed option stays. Google keeps running Claude on its Agent Platform, the route that earlier Claude on Vertex AI projects use (Anthropic's SDK client for it is still called AnthropicVertex). Google is the data processor there, under your existing Google Cloud agreements.
3. OpenAI models are coming to Model Garden. OpenAI's October 21 post says its frontier models and Codex workflows are coming to Model Garden, that the models will be bought through Google Cloud Marketplace, that customers with Google Cloud spend commitments can draw them down, that general availability is planned "in the coming weeks" and that Microsoft remains OpenAI's primary cloud partner.
4. Google's side of both deals. In its October 22 post, Google names the three OpenAI models in private preview, says preview customers can connect Codex CLI and Desktop to its platform for inference, and says OpenAI usage counts toward Flexible Savings Plans as well as commitments, with zero data retention options and Confidential Compute "coming soon". Its Claude post says both Claude options carry "the same pricing" and the same Google Cloud billing. Google calls itself the only hyperscale cloud offering all three leading closed-source model families, a point Techzine repeated in its October 21 report.

What Claude costs on Google Cloud

Anthropic's pricing page and Google Cloud's Agent Platform price list give the same standard rates per million tokens, read on October 23:

1. Claude Fable 5.1: $10 input, $50 output.
2. Claude Opus 5.5: $4 input, $20 output.
3. Claude Sonnet 5.5: $2 input, $10 output.
4. Claude Haiku 5.5: $0.10 input and $0.50 output for prompts up to 100,000 tokens, $0.50 and $2.50 above that.

Four details move the bill. Fast mode on Opus 5.5, which Anthropic says gives up to 2.5 times faster output, costs $8 input and $40 output per million tokens, and on Google Cloud it is listed only for the Anthropic-managed option. US-only inference through the inference_geo parameter adds a 1.1x multiplier on the Anthropic-managed option, the same as on Anthropic's API, while on the Google-managed option multi-region and regional endpoints carry a 10 percent premium over global ones. Anthropic meters usage and reports it to Google in Claude Consumption Units worth $0.01 each, so no separate Anthropic invoice arrives. And if you have a negotiated discount with Anthropic or Google, Anthropic's setup guide says to talk to your account representative before you switch, because discounts cannot be applied to past usage.

Anthropic managed or Google managed

Same models, same prices, same invoice. The difference, per Anthropic's comparison page and setup guide, is who runs the service and how fast features arrive:

1. Data processor. Anthropic, under its Commercial Terms, on the Anthropic-managed option, where some supporting services such as trust and safety process data outside Google Cloud. Google, under your existing Google Cloud agreements, on the Google-managed one.
2. Features. The Anthropic-managed option gets Claude Managed Agents (beta), code execution, Agent Skills, the advisor tool, the MCP connector, the Files API, web search, web fetch, batches and fast mode from day one. The Google-managed option gets new platform features once Google integrates them, plus VPC Service Controls, data residency in the US, EU and APAC, and links to BigQuery and Model Armor. Customer-managed encryption keys work on both.
3. Accounts and keys. The Anthropic-managed setup creates a new Anthropic organization for your Google Cloud project, separate from any Anthropic organization you already have, Claude Enterprise included. It accepts Google Cloud credentials only, with no API keys, and a zero data retention agreement on another Anthropic organization does not carry over.
4. Rate limits. Anthropic sets them on the Anthropic-managed option and measures them as its rate limits page describes, which our Claude API rate limits guide walks through; higher limits go through Anthropic's account team or sales. Google Cloud sets them on the Google-managed option.

What OpenAI on Google Cloud will cost

Google has not published Model Garden prices for OpenAI's models yet. OpenAI's own API price list, read on October 23 for prompts under 272,000 tokens, puts GPT-6 Astra at $10 input, $1 cached input and $50 output per million tokens, GPT-6.1 Sol at $2, $0.10 and $10, and GPT-6 Luna at $0.10, $0.01 and $0.50, with batch at half price and data residency at 10 percent extra. Those are the same input and output list prices as Claude Fable 5.1, Sonnet 5.5 and Haiku 5.5, tier for tier. Until Google prints its own numbers, treat OpenAI's list as the reference and check the Marketplace listing before you commit spend.

What it means for operators

The change is in procurement, not capability: these are the models you can already call directly. Where it pays off:

1. Companies with a Google Cloud commitment. Claude spend now counts toward it the way other Marketplace purchases do, per Anthropic's setup guide, and OpenAI spend will once it reaches your account. If your commitment is under-used, that is budget you had already promised Google.
2. Agencies building for clients. You can deploy into the client's own Google Cloud project: the client's IAM grants your team a role such as Claude User, the usage lands on the client's invoice, and nobody passes API keys around, because the Anthropic-managed option does not accept them.
3. Regulated work. The data processor choice is now explicit. If you are scoping AI intake for a law firm or AI calling for an insurer, decide early whether the client needs Google as processor with regional endpoints, or Anthropic's newest agent features.
4. Startups counting on credits. Anthropic's guide talks about commitments, not promotional credits, and our AI startup credits guide found that Google's program credits cover Gemini and Gemma rather than third-party models. Ask Google before you plan a Claude build on credits.

How to start: check which option you use today from the host in your request URLs (claude.googleapis.com is Anthropic managed, aiplatform.googleapis.com is Google managed), subscribe to the Anthropic-managed listing in Marketplace, enable claude.googleapis.com and anthropic-managed-claude.cloudpartnerservices.goog on each project, then activate the new Anthropic organization and grant roles. For OpenAI, request access through OpenAI's form and plan around general availability rather than the preview. If you want this wired into a working agent, see our AI automation agency service or hire an AI engineer.