Claude on Google Cloud gets betas the day they ship

On October 22, Anthropic made Claude Models on Google Cloud (Anthropic managed) generally available. Under this setup, Anthropic runs the Claude API itself, and Google Cloud handles identity, billing and committed spend. Because the API is Anthropic’s own, new features and betas reach Google Cloud customers on the same day they go live on the native Claude Platform.

Teams sign in with their existing Google Cloud IAM credentials, so they keep the roles and permissions they already manage, Anthropic said in its announcement. Usage appears on the single Google Cloud invoice and draws down the customer’s existing Google Cloud commitment, the spend a company has already pledged to Google by contract. The service has two endpoints for now: US and Global.

Who processes the data

The two Claude offerings on Google Cloud divide the work differently. In the new offering, Anthropic operates the service and acts as the data processor, the party that handles customer data on the customer’s behalf. Google provides the account layer around it: who may call the API, and which invoice the usage lands on. The existing Google managed offering stays available, and in that one Google is the data processor.

That split also decides the feature set. Calls go to Anthropic’s platform, so the Anthropic managed offering gets the full Claude API. For agents, that covers Claude Managed Agents (beta) for long-running agents without managing infrastructure, code execution for running Python and building visualizations inside an API call, computer use for operating a browser or desktop, Agent Skills for teaching Claude repeatable workflows, and the advisor strategy (beta), where an agent consults an advisor model. For data and tools, there is the Files API, citations, web search and web fetch, and the MCP connector (beta), which connects Claude to other tools without client code. Prompt caching and batch processing cover cost and latency.

Customers on this offering also get the Claude Console, Anthropic’s environment for writing prompts, configuring agents, running them on hosted environments, monitoring usage and costs, and managing access and security per workspace. The models available are Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, and Anthropic says new models will ship on this offering as they launch.

Fast mode on Opus 5.5 is generally available here as well. It is built for speed, with up to 2.5x faster output token generation, and Anthropic pitches it for rapid iteration, live debugging and interactive agents where output speed shapes the user experience. Input tokens cost $8 per million and output tokens $40 per million.

Which route to pick?

Anthropic matches each offering to a set of needs. The Anthropic managed offering is for customers who want day-zero access to every beta and generally available Claude Platform feature, with full parity with the native platform, while buying through Google Cloud. The Google managed offering remains the recommended choice for customers who need Google Cloud as the data processor, have strict regional data residency requirements, or want native Google Cloud integrations such as Model Armor or Agent Engine.

Anthropic describes Claude as the only frontier model available on all three of the largest cloud platforms: Bedrock on AWS, Vertex AI on Google Cloud and Foundry on Microsoft Azure. Google is also an investor. On April 24, it announced an investment of up to $40 billion in Anthropic, $10 billion of it immediately and up to $30 billion more tied to performance milestones. The new offering arrives six months later.

For teams that already have a negotiated price, the order of steps matters. Anthropic asks customers with an existing discount to contact their Anthropic or Google Cloud account executive before they start on the Anthropic managed offering, so the discount is applied correctly. Discounts cannot be applied retroactively to usage.