Claude Fable 5 is now available on Google Cloud's Gemini Enterprise Agent Platform

Claude Fable 5 lands in Google Cloud’s Model Garden

Claude Fable 5, Anthropic’s first Mythos-class model made generally available, is now accessible through Model Garden on the Gemini Enterprise Agent Platform. For teams already running workloads on Vertex AI, that means you can add Fable 5 to your projects using the same IAM controls and billing setup you already have in place, without opening a separate account with Anthropic.

This is a meaningful access change. Previously, using Claude at this capability tier required either a direct Anthropic API contract or access through Anthropic’s limited early programme. Now it sits alongside Gemini and other third-party models in Model Garden, discoverable and deployable from inside your GCP project.

What is Claude Fable 5, exactly?

Fable 5 is Anthropic’s general-release version of its Mythos-class frontier models. Anthropic initially kept Mythos-level capability restricted to a small group of cyber defenders and critical infrastructure providers under Project Glasswing. The Fable 5 release marks the point where Anthropic considered its safety safeguards robust enough for broader use.

Fable 5 includes safety classifiers that the raw Mythos 5 model does not. Those classifiers can decline requests, returning a stop_reason: "refusal" in the API response rather than an error code, so your integration needs to handle that as a valid response path if you’re building on top of it.

On capability: it scores 80.3% on SWE-Bench Pro (versus 69.2% for Opus 4.8 and 58.6% for GPT-5.5). It shares a one-million-token context window with Claude Mythos 5 and supports up to 128k output tokens per request. It’s designed to run in long-horizon agent harnesses, working for extended periods across multiple stages, delegating to sub-agents, and reviewing its own output.

Pricing is $10 per million input tokens and $50 per million output tokens, which Anthropic notes is less than half the price of Claude Mythos Preview.

How access actually works on Vertex AI

The setup is straightforward but does involve a few steps beyond a standard model deployment. Here’s the honest picture:

You do not need a standalone Anthropic API subscription. Billing runs through your Google Cloud account in the normal way.

You do need to complete a consent process. Specifically, you must:

Accept Anthropic’s terms and Google Cloud Marketplace terms of service via Cloud Marketplace
Accept a Private Offer in Cloud Marketplace (this is required once per Billing Account)
Consent to the Advanced AI Safety Addendum in Model Garden (required once per project)
Enable prompt and response sharing with Anthropic for each model location

That last point is worth noting. Fable 5 cannot run under zero data retention. Anthropic’s policy for Mythos-class models requires that prompts and outputs are retained for 30 days for trust-and-safety review, after which they are automatically deleted. The data stays within the GCP environment and Anthropic staff cannot access it unless something is flagged for potential serious harm or you request it in writing.

For most enterprise teams, this process is a one-time setup rather than an ongoing overhead, but it is worth factoring in if your organisation has strict data handling requirements that assume zero retention.

Once enabled, the Vertex AI model ID is claude-fable-5 with no date suffix. One practical note for teams getting set up: without the data sharing configuration in place, API calls will return a 403 PERMISSION_DENIED error. This is resolved by enabling data sharing for the Anthropic publisher via the setPublisherModelConfig API.

What this means for your GCP environment

The practical benefit is consolidation. If your team is already using Vertex AI, pulling in Fable 5 means you get:

Access control through IAM using the same roles and permissions as the rest of your GCP services
Unified billing under your existing Google Cloud account, so Claude usage appears alongside your other infrastructure costs
Cloud Logging integration for audit trails and monitoring
No separate credential system for developers to manage

That last point matters more than it sounds. One of the friction points with adopting third-party AI providers is the proliferation of API keys, separate accounts, and billing relationships across a team. Running Fable 5 through Vertex AI removes that friction for teams that are already Google Cloud-native.

Where it fits in your stack

Fable 5 is positioned for demanding workloads: complex multi-stage knowledge work, financial analysis involving dense PDFs and tables, software engineering at scale, and long-running agent tasks where you hand off a project and review completed work rather than supervising each step.

If you’re running Claude Code or another agent harness on Vertex AI, Fable 5 is worth evaluating as the underlying model, given its benchmark performance on software engineering tasks and its design for extended autonomous operation.

It’s worth noting that Fable 5’s retirement date is no sooner than June 8, 2027, so if you’re building production integrations, you have a defined support horizon to plan against.

A note on the broader Gemini Enterprise Agent Platform context

Model Garden is now part of the Gemini Enterprise Agent Platform, which is Google Cloud’s rebranded and expanded evolution of Vertex AI. It brings model selection, agent building, orchestration, and governance tooling under a single platform umbrella. Fable 5’s availability here is part of a broader pattern of Google Cloud positioning the platform as a neutral place to run frontier models from multiple providers, not just Google’s own.

Claude, Grok, and Mistral are all available as managed APIs through Model Garden. Fable 5 is the newest and highest-capability addition to that lineup.

Sources: Google Cloud Gemini Enterprise Agent Platform release notes, Claude Fable 5 on Agent Platform enablement docs, Anthropic Claude Fable 5 announcement