On October 22, 2026, Anthropic's Claude release notes announced Anthropic-managed Claude on Google Cloud: the Claude API, operated by Anthropic, reached through your Google Cloud project. Our read: the models and base token prices stay the same, and what changes is the contract, the invoice, the SDK client (AnthropicGoogleCloud, in beta, versus AnthropicVertex) and the residency options. That's a budget decision, not a reason to move production by itself. For the wider pattern, see AI model releases and pricing: what changes for engineering teams.
What launched on October 22
The release notes put it in four facts. Anthropic operates the service. Google Cloud IAM controls access. Google Cloud Marketplace handles billing. And fast mode is available for Claude Opus 5.5 on the new route.
Google's launch post adds the contract framing. Anthropic runs the stack as the data processor on Google Cloud infrastructure under Anthropic's terms, while you authenticate with Google IAM, keep Cloud Audit Logging and consolidate spend under your Google Cloud billing and eligible commitments. The older route, where Google operates Claude and acts as processor under your existing Google customer agreements, stays. Claude's docs now call it Google managed. It's the route you'd know as Claude on Vertex AI.
Google's post also says both options deliver "the same pricing." Hold that thought until the residency section, because the residency premiums complicate it.
Three routes, one table
No vendor page puts these side by side. Cells say "not stated" where the docs don't answer.
| What changes | Direct Anthropic | Anthropic-managed on Google Cloud | Google-managed |
|---|---|---|---|
| Terms you accept | Anthropic's | Anthropic's Commercial Terms and Usage Policy, accepted at setup | Google's existing customer agreements |
| Data processor | Not covered here | Anthropic (some trust and safety processing happens outside Google Cloud) | Google Cloud |
| Invoice | Anthropic bills you (our inference) | Google Cloud bill, in Claude Consumption Units of $0.01; no separate Anthropic invoice | Google Cloud bill |
| Google commitment drawdown | No (our inference) | Counts "the way other Marketplace purchases do" | "Eligible commitment drawdown," per Google |
| Authentication | API keys | Google Cloud credentials only | Google Cloud credentials |
| Rate limits | Anthropic | Anthropic | Google Cloud (limits and quotas) |
| Residency | Global or US-only | Global or US-only | US or EU multi-region; single-region endpoints |
| Zero Data Retention | Not covered here | On request via your Anthropic rep | Data handling governed by Google Cloud |
| Model lifecycle dates | Anthropic's | Not stated on the Anthropic-managed page | Set by Google, can differ from Claude API |
Does the spend draw down your Google commitment?
For Anthropic-managed, Claude's docs say usage counts toward a Google spend commitment the way other Marketplace purchases do. Google's Marketplace drawdown page, last updated October 22, says in substance that up to 100% of eligible Marketplace spend may count toward a minimum commitment, with listed exclusions (one is Marketplace services not running on Google Cloud), and that Google can change the list at its sole discretion.
Note the words "may" and "eligible." Whether your Claude spend counts at 100%, or is capped, depends on your agreement, and nothing public tells you. Get the cap in writing. Our judgment: direct Anthropic spend doesn't draw down a Google commitment, since Anthropic bills it separately.
We covered the same question for another vendor in the same drawdown question for OpenAI models. The logic carries over: you're burning committed spend you'd owe anyway, and you're not getting a cheaper token.
Which contract and DPA applies
Anthropic-managed runs under Anthropic's terms, with Anthropic as processor. Google-managed runs under Google's agreements, with Google as processor and governing data handling. If your legal team signed one and not the other, you'll be signing a new document either way.
What the docs don't say is whether your existing Anthropic master agreement or DPA covers the new organization. No page we read addresses it. Don't assume it does, and ask your account team to answer in writing before any traffic moves.
Residency and encryption
This is where the Google blog's "same pricing" line needs a footnote.
Google-managed offers multi-region us and eu endpoints plus single-region endpoints. Anthropic-managed offers global or us only: every workspace uses us as its workspace geo, and the Console shows the settings read-only. If you need EU-only inference, Anthropic-managed can't give it to you today. Google-managed can.
Both routes charge about 10% for pinning geography, just under different names. Anthropic's data residency docs price US-only inference (inference_geo: "us") at 1.1x the standard rate on the Claude API and on Anthropic-managed Google Cloud, for Claude 4.6 and later. The Google-managed page says multi-region and regional endpoints cost 10 percent more than global ones (for Sonnet 4.5 and later). So Google's "same pricing" holds on the global endpoint, and both routes add a premium once you pin geography. One more constraint: regional endpoints serve Claude Sonnet 4.6 and earlier only, so newer models need global or multi-region.
Customer-managed encryption keys are muddier. Claude's Anthropic-managed page says CMEK is available with Google Cloud KMS keys. Google's blog lists CMEK only under the Google-managed route. We're reporting the discrepancy, not picking a winner. If CMEK is a requirement, get it confirmed for your route.
Feature lag, in both directions
The Google blog frames Anthropic-managed as the way to get new features sooner, and the Claude docs confirm both the direction and the size of the gap. Google-managed doesn't support the Files API, URL sources for images and documents, code execution, web fetch, advisor, Agent Skills, the MCP connector, programmatic tool calling, the Message Batches API, the Admin, Compliance and Usage and Cost APIs, Claude Managed Agents, or server-side fallback. Google's own post says new platform features roll out as they're integrated into Google-managed infrastructure.
The reverse exception is small: Anthropic-managed has no API keys, no MCP tunnels, no pre-built self-hosted sandbox workers and no memory stores on self-hosted environments.
What switching costs
You can't convert an existing Anthropic organization, including one you use with the API directly. Setup creates a new organization, separate from any you already have, Claude Enterprise included. Then:
- API keys, workspaces and Console settings don't carry over.
- Files, skills, batches and Managed Agents resources stay in the old organization.
- A Zero Data Retention arrangement on another organization doesn't carry over.
- Discounts can't be applied to past usage, so talk to your rep before you start.
- The SDK client for the route is in beta.
The good news is that the two organizations are independent and can serve traffic in parallel, and the old one keeps its current billing. You can pilot without cutting over. Moving from Anthropic-managed to Google-managed is mostly a client swap, plus a model-ID change for models before 4.6. The reverse needs the new-organization setup above.
What this means for your 2027 AI budget
Decide the contract route before you write the Claude line into the 2027 budget, because the same tokens will land on a different invoice under a different processor.
Our judgment, by situation:
- A Google commitment with unspent room and no EU-only need: pilot Anthropic-managed, and keep production where it is until the answers below are in.
- EU-only inference: Google-managed is the only route offering it.
- Already on direct with negotiated discounts or ZDR: ZDR doesn't carry over and the docs don't say whether discounts transfer, so stay put unless the commitment burn-down is worth more.
- Dependence on Files, Skills or Batch: Google-managed doesn't support them.
Put four questions to your account teams:
- What is the Marketplace drawdown cap for Claude, in writing?
- Does our existing Anthropic MSA and DPA extend to the new organization?
- When would a ZDR arrangement be in place, and is that date committed?
- Do negotiated discounts transfer?
To size the spend you're routing, see what AI coding tools cost per developer.
FAQ
What is Claude Models on Google Cloud (Anthropic managed)?
It's the Claude API operated by Anthropic and reached through your Google Cloud project. Google IAM controls access and Google Cloud Marketplace handles billing. It was announced in Claude's release notes on October 22, 2026.
Is it the same as Claude on Vertex AI?
No. The Vertex-style route is now called Google managed: Google operates the service under its own agreements and acts as processor. The new route is run by Anthropic under Anthropic's terms. They support different features and residency options.
Does Anthropic-managed Claude count toward my Google Cloud commitment?
Claude's docs say usage counts the way other Marketplace purchases do. Google's drawdown rules apply, and any cap sits in your own agreement. The docs don't state it, so ask your Google account team.
Who is the data processor?
On Anthropic-managed, Anthropic. On Google-managed, Google Cloud. Some Anthropic supporting services, such as trust and safety systems, process data outside Google Cloud.
Can I keep EU data in the EU?
On Google-managed, yes, with a multi-region eu endpoint. On Anthropic-managed, inference is global or us only.
Can I run both routes at once?
Yes. Claude's docs say the new organization and your existing one are independent and can serve traffic in parallel.
Sources
- Claude Docs: Release notes overview (October 22, 2026)
- Claude Docs: Claude Models on Google Cloud (Anthropic managed)
- Claude Docs: Claude Models on Google Cloud (Google managed)
- Claude Docs: Data residency
- Google Cloud blog: Expanding Claude availability on Google Cloud
- Google Cloud docs: Marketplace commitment drawdown