Anthropic-Managed Claude Platform Goes GA on Google Cloud: Full Features, Google Billing

What actually changed

On October 22, 2026, Anthropic announced that Claude Models on Google Cloud (Anthropic managed) is generally available (GA). In one sentence: Google Cloud customers can now get the full Claude Platform without waiting for Google to integrate each new feature into its own managed version. Google Cloud confirmed the same thing the next day, October 23, in its official blog: the Anthropic-managed deployment option is GA, alongside the existing Google-managed option.

The key point is that this is not another new model launch. It is a gap closed in how Claude is deployed. Both options offer the same frontier models — Anthropic lists Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5, and promises that new models will ship on the Anthropic-managed option as they launch. The real differences are who operates the service, who acts as data processor, when new features arrive, and whose terms you sign.

For teams already running on Google Cloud, there is another practical point: billing does not split. Anthropic says you keep your existing Google Cloud IAM credentials, billing goes through a single Google Cloud invoice, and Claude usage draws down your existing Google Cloud commitment. Google's blog says the same thing: consolidated billing and eligible commitment drawdown. In the vendors' own words, at least, you do not need a separate procurement and permissions setup just to get the full Claude Platform.

Two routes, one comparison table

Google Cloud now offers two ways to use Claude. The table below combines what Anthropic's announcement and Google Cloud's blog say; where they agree, you can read it with confidence, and where only one side mentions something, the text after the table calls it out.

Dimension	Anthropic managed (new GA)	Google managed (existing option)
Operator and data processor	Anthropic operates the service; Anthropic is the data processor	Google Cloud is the data processor
Terms	Anthropic's Commercial Terms	Existing Google Cloud customer agreements
Model availability	Day zero, as models launch	Day zero, as models launch
New platform features and betas	Same day as the native Claude Platform	Roll out once integrated into the Google-managed infrastructure
Identity and permissions	Google Cloud IAM credentials	Google Cloud native identity
Billing and commitments	Single Google Cloud invoice; draws down eligible commitments	Standard billing; draws down eligible commitments
Audit and controls	Cloud Audit Logging retained (per Google's blog)	VPC Service Controls, Customer-Managed Encryption Keys (CMEK)
Data residency	US and Global endpoints currently available	Data residency in the US, EU, and APAC
Native integrations	Full Claude Platform feature set	Model Armor, Agent Engine, BigQuery, and other Google Cloud native integrations

The row most easily misread is "model availability." Google's blog states explicitly that both routes get new models on day zero. Choosing Google managed does not put you behind on the models themselves; what lags is everything around the models — Managed Agents, skills, code execution, and future betas. Conversely, choosing Anthropic managed is not an "upgrade." It trades the data processor and contracting party for Anthropic. That is a trade-off, not a promotion.

What the "full platform" actually includes

Anthropic's feature list is more specific than "the whole API family," and it is worth breaking down by use.

First, agent capabilities. Claude Managed Agents (beta) targets long-running agents without managing infrastructure; Agent Skills teaches Claude repeatable workflows so results stay consistent; Advisor Strategy (beta) lets an agent consult an advisor model mid-task, adding a layer of intelligence review. Computer Use lets Claude operate a browser or desktop to complete tasks — useful for automation, but also a capability whose permissions and audit trail need thinking through before you look at the demo.

Second, connecting to systems and knowledge. The Files API uploads and references documents across conversations; Citations grounds answers in source documents; Web Search and Web Fetch add current, real-world data; the MCP Connector (beta) connects Claude to other tools without writing client code. For teams building internal assistants, support tools, or developer aids, this group often matters more than model specs — it decides whether Claude can read your own data and whether its answers carry sources.

Finally, cost and performance management. Prompt Caching cuts cost and latency on repeated context; Batch Processing handles high-volume asynchronous work. Anthropic also includes Claude Console, its development environment: write prompts, configure agents, run them on hosted environments, monitor usage and costs, and manage access and security per workspace. This was previously the full native Claude Platform experience; it is now available through Google Cloud too — provided you pick the Anthropic-managed route.

Fast Mode and pricing: only vendor-stated numbers

Shipping alongside the GA is another announcement: Fast Mode on Claude Opus 5.5 is also generally available to Anthropic-managed customers. Per Anthropic, Fast Mode prioritizes speed, offering up to 2.5x faster output token generation, for rapid iteration, live debugging, and interactive agents where output speed improves the experience. Note "up to" — that is a vendor claim, not a measured guarantee, and real speedups will vary by task and load.

On pricing, Anthropic's announcement gives a vendor-stated figure: input and output tokens for Opus 5.5 Fast Mode are $8 and $40 per million, respectively. Neither primary announcement contains a full price list for the other models under this option, nor an item-by-item comparison with standard mode. If you are budgeting, treat this figure as a starting point and check Anthropic's official pricing page for the models and modes you actually use — do not take this one pair of numbers as the price of all Claude models.

Google's blog is more general: both routes offer "the same pricing." The sensible reading is that the same model does not change price just because of the route — not that every model and mode has a single price. Budget against Anthropic's price list; the route itself is not a discount.

Who should pick which route

If your team already uses agents, skills, and code execution on the native Claude Platform, or your roadmap depends on beta features being available immediately, Anthropic managed is the more direct choice: features arrive at the same time as the native platform, with no integration queue. Likewise, if your workflow already revolves around Claude Console, the Files API, and Citations, this route keeps those capabilities without downgrades or substitutes.

If your organization has hard requirements on who the data processor is — a contract, compliance rule, or customer commitment naming Google Cloud — or strict regional data residency needs (for example, data that must stay in the EU or APAC), Google managed remains the option Anthropic itself recommends. Anthropic's own words: customers who need Google Cloud as the data processor, have strict regional data residency requirements, or want native integrations like Model Armor or Agent Engine should stay with Google managed. Google's blog lists the same reasons and adds CMEK and VPC Service Controls.

There is also a common middle case: you already run Claude on Google managed, the models work fine, and you are unsure whether to switch. The decision should not rest on the feature list alone, but on whether you actually lack the platform features that have not yet been integrated. If a specific beta or platform capability is missing, switching has a clear payoff; if all you want is a newer model, both routes already deliver it on day zero, and switching for its own sake buys nothing.

Five things to check before switching

1. Do not assume discounts carry over. Anthropic's announcement is explicit: customers with an existing discount should contact their Anthropic or Google Cloud account executive before starting with the Anthropic-managed offering, to ensure discounts are applied correctly. Discounts cannot be applied retroactively to usage. There is no exception stated — agree the discount first, then generate usage, not the other way around.
2. The contracting party changes. Anthropic managed runs under Anthropic's Commercial Terms; Google managed runs under existing Google Cloud customer agreements. If your legal or procurement process must approve the contracting entity and data-processing terms, clear that before engineers switch anything on in the console.
3. The data processor changes. The flip side of the same point: the data processor moves from Google to Anthropic. That affects privacy assessments, customer commitments, and internal compliance documents, which should be listed for update before the switch, not after.
4. Confirm the endpoint coverage. Anthropic's announcement says US and Global endpoints are currently available. If your business has specific regional requirements, check them one by one against Google managed's US, EU, and APAC residency options — do not assume a "Global endpoint" means data can land in any region.
5. Check the native integrations you rely on. If your workflow depends on Model Armor, Agent Engine, or BigQuery integrations, confirm before switching whether and how those integrations work under Anthropic managed. Neither announcement provides a line-by-line list of "all native integrations available on the new route," so this cannot be assumed.

Do not confuse this with another VibeFix article

A boundary is worth drawing here. VibeFix previously published an article on OpenAI models arriving on Google Cloud (slug: openai-models-google-cloud-model-garden-october-2026), about OpenAI's models entering Google Cloud's Model Garden, with Codex billable through Google. That article is about OpenAI and GPT-series models. This article is entirely different: its subject is Anthropic, and its product is the Claude Platform's Anthropic-managed deployment option on Google Cloud. The only thing the two share is the platform — Google Cloud — and the pattern of "model vendor plus cloud billing." Vendor, models, deployment options, and data-processor arrangements are all different, and facts from one must not be cited for the other.

The boundary matters because the two stories landed close together in time and their headlines look alike. If you encounter both while searching for which large models are available on Google Cloud, the vendor name is the fastest disambiguator: the OpenAI piece is about Model Garden; the Anthropic piece is about Claude Platform's two deployment routes.

What is still unclear

Frankly, the two primary announcements do not answer every practical question. The following points could not be confirmed from the primary sources opened for this article, and are listed so readers do not fill them in by guesswork.

• Whether there is an in-place migration path from Google managed to Anthropic managed. The announcements say the new option is GA and who each option suits; neither describes how existing users migrate, whether usage history and configuration carry over, or whether switching involves downtime.
• The full price list beyond Opus 5.5 Fast Mode. As noted above, Anthropic's announcement gives only the Fast Mode figures, and Google's blog only says pricing is the same. The full list for other models must be checked on Anthropic's pricing page; this article does not invent it.
• Reseller or partner-channel terms. Neither primary announcement opened for this article mentions reseller restrictions or channel-specific terms. This article therefore draws no conclusion about resellers; if you buy Google Cloud through a partner, your own contract governs.
• What "Global endpoint" covers in detail. Anthropic's announcement lists US and Global endpoints without breaking down which regions the Global endpoint includes, or the latency and compliance implications of each choice.
• Integration timelines for specific features on Google managed. Google's blog says only that features roll out "once integrated," with no per-feature schedule. Teams waiting on a specific feature should follow that feature's later official updates rather than extrapolating from this article's publication date.

What this means for day-to-day development

For working developers, the most immediate effect is not a new console entry but a tidier workflow. Previously, a team that wanted both Google Cloud identity and billing and the full Claude Platform often juggled two processes: permissions managed on one side, agent configuration and usage monitoring on the other. Under Anthropic managed, as described by the vendors, identity stays in Google Cloud IAM, agent development, execution, and monitoring live in Claude Console, and billing lands on a single invoice. That removes process friction, not a model capability gap — it should be evaluated as a workflow gain, not a benchmark win.

In agent development, the combination worth noting is Managed Agents, Skills, and code execution together. Managed Agents covers the infrastructure burden of long-running tasks, Skills addresses inconsistent output on repeated tasks, and code execution lets the model run Python, build visualizations, and analyze data within a single call. Together they cover "runs continuously, runs consistently, and can do the math." None is novel on its own; together they mean a team does not need separate hosting or glue code for each step. That is still not turnkey: Skills must be built from your own workflows, and the sandbox boundaries, data access permissions, and result validation for code execution must be designed to your own security requirements.

On the knowledge side, the Files API, Citations, Web Search, and Web Fetch combination decides whether an internal assistant can be trusted. Being able to reference internal documents, anchor answers to sources, and supplement with current external information when needed is what moves an answer from "plausible" to "traceable." The MCP Connector's beta status is a reminder that convenience and stability are still converging: interfaces and behavior may change, so production systems should validate it on non-critical paths before depending on it.

Cost management also needs a different way of counting. Prompt Caching genuinely saves money where repeated context is large — but only if your tasks really do repeat large context. Batch Processing fits bulk work that can wait, not latency-sensitive interaction. Fast Mode trades for output speed, not better reasoning quality: real-time debugging and interactive agents benefit, while offline jobs that can run slowly gain little from paying a speed premium. In short, the GA delivers a set of dials to combine, not a single default-best configuration, and teams should test each against their own task types rather than copying the announcement's suggested scenarios wholesale.

One organizational effect is easy to overlook: when the data processor and contracting party change, internal documents, vendor lists, and security assessment records all need updating. That work creates no new features, yet it determines whether a switch clears internal approval. Listing the documents to update and naming who approves them, ahead of time, often takes longer than the technical migration — and is more often underestimated.

Finally, a note on how to read this kind of news. Cloud-and-model-vendor combinations are multiplying, and "same model, two hosting styles" will likely appear again. The reliable way to read such stories is to separate three questions: is the model new, where does the feature-arrival gap sit, and who became the contracting party and data processor? Keep those apart, and phrases like "generally available" and "full platform" will not be misread as "a stronger model" or "a blanket upgrade" — and the selection debate can land faster on the costs you will actually pay and the responsibilities you will actually carry.

One-line verdict

The substance of this GA is that Anthropic has made "the full Claude Platform plus Google Cloud identity and billing" a formal option, rather than forcing Google Cloud users to choose between a complete feature set and staying inside the Google system. Each route now has a clear trade-off: pick Anthropic managed for feature-arrival speed; stay with Google managed for Google as data processor and native integrations. Before choosing, run through the five checks — discounts, contracting party, data processor, endpoints, and native integrations — it saves far more trouble than demoing first and fixing process later.