Anthropic Brings Claude Directly to Google Cloud With Same-Day Feature Access
Anthropic now runs Claude directly on Google Cloud with day-zero feature parity, GCP billing, IAM, and commitment drawdown, no Vertex AI required.
- Claude Models on Google Cloud (Anthropic managed) is now generally available with full Claude Platform parity.
- New features and betas ship the same day as the native Claude Platform, closing the Vertex AI lag.
- Uses existing Google Cloud IAM, invoicing, and draws down GCP commitments; Anthropic is the data processor.
- Fast Mode on Opus 5.5 ships GA at $8 input and $40 output per million tokens, up to 2.5x faster.
- Includes Managed Agents, code execution, Skills API, computer use, Advisor, MCP connector, prompt caching, and the Claude Console.
- Google managed path remains for data residency, Model Armor, and Agent Engine integrations.
Anthropic has made Anthropic-managed Claude generally available on Google Cloud. The service exposes the native Claude Platform while using Google Cloud IAM for authentication and Google Cloud for billing and procurement.
Teams can purchase the service through the Google Cloud Marketplace, charge token usage to their existing invoice, and apply committed Google Cloud spend. Access requires neither a separate Anthropic contract nor a new credential system.
Native features arrive on launch day
Before this release, Google Cloud customers generally accessed Claude through Vertex AI. In that model, Google serves as the data processor, and Anthropic features follow a separate Vertex release schedule. Vertex also provides IAM integration, VPC controls, logging, monitoring, regional endpoints, and other controls used in regulated deployments.
Because Anthropic operates the new service, features and betas arrive alongside their release on the native Claude Platform. Capabilities available at launch include:
- Fast Mode for Claude Opus 5.5: Up to 2.5 times faster output-token generation, according to Anthropic, at $8 per million input tokens and $40 per million output tokens.
- Code execution: Python execution, chart generation, and data analysis within an API call.
- Agent Skills API: Reusable instructions and workflows that teams can package for Claude.
- Claude Managed Agents beta: Hosted infrastructure for long-running agent workloads.
- Computer use: Browser and desktop interaction for tasks that require graphical interfaces.
- Advisor strategy beta: A primary agent can consult a higher-capability model during a task.
- Platform tools: Prompt caching, batch processing, the Files API, Citations, web search, web fetch, and the MCP connector beta.
Model coverage includes Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5, with future models planned for launch alongside the native platform. Anthropic describes Fable 5.1 as its most capable generally available model for long-running, asynchronous jobs, making it a natural fit for managed agent workflows.
One cloud, two control planes
Google Cloud customers can now choose between Anthropic-managed Claude and the existing Google-managed Vertex AI route. The options differ in service operation, data processing, feature cadence, regional coverage, and access to platform-specific tools.
| Dimension | Anthropic managed | Google managed through Vertex AI |
|---|---|---|
| Data processor | Anthropic | |
| Feature cadence | Same-day access to native Claude releases | Separate Vertex AI rollout schedule |
| Best fit | Native Claude APIs, betas, and the Claude Console | Strict data residency, Model Armor, and Agent Engine |
| Authentication and billing | Google Cloud IAM, invoice, and commitment drawdown | Google Cloud IAM and invoice |
| Launch endpoints | United States and global | United States, Europe, multiregion, and global |
The processor designation affects contracts, data-governance reviews, and compliance architecture. Anthropic processes API traffic on the Anthropic-managed route even though Google supplies authentication and billing. Google processes traffic sent through Vertex AI.
Anthropic-managed customers also receive the Claude Console, Anthropic’s development environment for writing prompts, configuring agents, running them in hosted environments, and monitoring workspace usage, cost, and access. This gives teams Anthropic’s first-party development and observability tools within a Google Cloud procurement arrangement.
Anthropic splits operations from sales
Claude already reaches cloud customers through AWS Bedrock and Microsoft Foundry, while its Google relationship has expanded since Claude 3 entered Vertex AI Model Garden. Under the new arrangement, Anthropic operates the service on Google infrastructure, and Google manages authentication, billing, procurement, and committed-spend accounting.
Anthropic’s launch materials cite feature cadence and cost management as the main adoption drivers. Abacus.AI’s chief technology officer highlights launch-day feature access. Triple Whale, which runs its Moby ecommerce agent on Claude, says prompt-caching behavior carries over unchanged while billing moves through Google Cloud. Cache consistency matters for applications whose unit economics depend on high cache-hit rates.
From beta access to committed spend
Teams with existing Google Cloud commitments can adopt native Claude capabilities without creating a separate commercial and identity stack. They can now:
- Deploy Claude Managed Agents under existing Google Cloud IAM policies.
- Use Advisor strategy and the Agent Skills API as Anthropic releases them.
- Retain native prompt-caching behavior for workloads sensitive to cache-hit rates.
- Apply committed Google Cloud spend to Claude token usage.
- Use the Claude Console for prompt development, agent configuration, and workspace monitoring.
Production adoption still requires a review of commercial terms, data handling, and regional controls:
- Discounts: Existing Anthropic or Google Cloud discounts require coordination with the relevant account team. They cannot be applied retroactively to usage on the new service.
- Data residency: The Anthropic-managed service launches with United States and global endpoints. Vertex AI offers additional European and multiregion options.
- Google-native controls: Workloads that depend on Model Armor or Agent Engine should remain on the Vertex AI route.
- Data processing: Security and legal teams should review Anthropic’s role as processor before moving traffic from Vertex AI.
Comments
