Same models, same price, different processor
Google Cloud now sells Claude in two forms, and Google is explicit that price does not separate them. Its announcement states that both options deliver the same frontier models on day zero, the same pricing and the same Google Cloud billing. That is Google's published position on standard pricing; negotiated discounts, regional charges and eligible commitments remain each customer's own to verify. So the decision a buyer actually faces is elsewhere.
On the new Anthropic-managed route, Anthropic operates the stack and acts as the data processor, under Anthropic's terms, while the customer signs in with Google Cloud IAM, keeps Cloud Audit Logging and consolidates spend under its Google Cloud billing and eligible commitments. In return, the customer gets the platform itself, not a cloud port of it: Managed Agents, skills and code execution from launch day, and every later feature and beta on the day it reaches Anthropic's own platform.
On the existing Google-managed route, Google operates the service as data processor under the customer's existing agreements. Google lists what that route carries and the new one, in its announcement, does not: VPC Service Controls, customer-managed encryption keys, and data residency in the US, EU and APAC, plus native integrations such as BigQuery and Model Armor. Its cost is timing. New platform features reach it over time, as Google integrates them.
That is a genuine trade, and it will be settled by compliance teams rather than by developers. A team that needs European residency or its own keys does not have a free choice of the faster feature track. A team that wants Managed Agents through this newly announced Google Cloud offering on day one must accept Anthropic as the data processor to get them. Neither announcement spells out the data-retention terms that would complete the picture; what each specifies is who operates the service, who processes the data under which agreement, and which controls each route advertises.
Check the endpoint list before the feature list
Anthropic's own announcement gives the Anthropic-managed service two endpoints: US and Global. It does not list an EU or APAC endpoint. Set beside Google's residency list for the Google-managed option, that is the practical map of this launch. Full features are available now, in the places Anthropic lists. Regulated workloads pinned to a region belong, for the moment, on the other route, where the features arrive later.
Anthropic also states the limit in the other direction plainly. In its words, the Google-managed offering “remains the best choice” for customers who need Google Cloud as data processor, have strict regional data residency requirements, or want native Google Cloud integrations such as Model Armor or Agent Engine. When the vendor selling the new route names the customers who should not take it, that sentence deserves more weight than the launch headline.
The discount clause is the expensive sentence
One instruction in Anthropic's announcement has a date attached to it in effect, even though none is printed. If you have an existing discount, contact your Anthropic or Google Cloud account executive before getting started with the Anthropic-managed offering, to ensure your discounts are applied correctly. Discounts cannot be applied retroactively to usage.
Read at face value, a customer who starts before confirming the applicable discount risks having that initial usage billed without the expected reduction. The announcement does not explain whether separate billing adjustments or credits may be available afterwards, so the safe reading is the literal one: the discount will not reach back over usage already run. That is a small clause with a direct bill attached, and it is the easiest part of this launch to get wrong, because nothing in the signup flow described in either announcement pauses to enforce it.
The rest of the buying decision can be revisited. Processor choice, endpoints and feature track can be weighed over a procurement cycle. Teams with negotiated rates should confirm the effective date and eligible usage with their account executive before enabling the service, and treat that call as step one, not as paperwork to finish later.
Sources
- [1] Anthropic, “Claude Platform is now available on Google Cloud”, claude.com, published 2026-10-22 (general availability of Claude Models on Google Cloud (Anthropic managed); the full feature list, including Claude Managed Agents (beta), code execution, computer use, Agent Skills, Advisor strategy (beta), Files API, Citations, web search and web fetch, MCP connector (beta), prompt caching, batch processing and the Claude Console; Google Cloud IAM authentication, single-invoice billing and commitment retirement; US and Global endpoints; the model list — Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5; Fast Mode on Opus 5.5 at $8 input / $40 output per million tokens, with up to 2.5x faster output token generation advertised; Anthropic operates the service and acts as the data processor; and the warning that discounts cannot be applied retroactively to usage)Read source
- [2] Google Cloud, “Expanding Claude availability and deployment choice on Google Cloud”, cloud.google.com blog, dated 2026-10-23 (both options deliver the same frontier models on day zero, the same pricing and the same Google Cloud billing — Google's published position; Anthropic operates the stack as data processor under Anthropic's terms on the new route, with Google Cloud IAM, Cloud Audit Logging and consolidated billing; Google as data processor under existing customer agreements on the Google-managed route, with VPC Service Controls, customer-managed encryption keys (CMEK), data residency in the US, EU and APAC, and BigQuery and Model Armor integrations, with new platform features rolling out over time)Read source