-NoNewline

Claude Fable 5 API: Model ID, Availability, Refusal, and Fallback

Claude Fable 5 is Anthropic's most capable generally available model (GA June 9, 2026), available on the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. It uses a safety classifier that returns HTTP 200 with stop_reason=refusal instead of an error code when a request is blocked — this requires dedicated monitoring and fallback logic.

For: Developers integrating Claude Fable 5 via API · Claude Code users needing Fable 5 (requires v2.1.170+) · Teams comparing Fable 5 across cloud providers · Engineers debugging Fable 5 refusal or 400 errors

What This Is

Claude Fable 5 is Anthropic's first publicly released Mythos-class model, positioned above the Opus tier. It targets long-running, complex, asynchronous agentic tasks. Core specs: 1M token context window, 128K max output per request, $10/M input tokens, $50/M output tokens, 90% discount on prompt cache reads.

Key difference from Opus: Fable 5 ships with safety classifiers covering offensive cyber work, biology, and reasoning extraction. When a request triggers a classifier, the API returns HTTP 200 with stop_reason=refusal rather than an HTTP 4xx error. This means monitoring systems built around 4xx/5xx will silently miss these events.

Refusal fallback: On the native Claude API and Claude Platform on AWS, a beta fallbacks parameter automates retry to Opus 4.8 server-side. On Amazon Bedrock, Vertex AI, and Microsoft Foundry, the fallbacks parameter is not available — you must implement client-side retry logic using the Anthropic SDK refusal-fallback middleware (Python, TypeScript, Go, Java, C#).

Claude Code: Fable 5 support requires Claude Code v2.1.170 or later. Older versions will not show Fable 5 in the model picker.

**Why Claude Fable 5 Is Unavailable**

If Fable 5 does not appear in /v1/models or returns an error, the cause is usually one of the following:

- **Model missing from /v1/models** — the account has not been granted access yet, or the model is not yet available in the account's region or tier. - **Wrong model ID** — using the Claude API format (claude-fable-5) on a platform that requires a different format (e.g., Bedrock requires anthropic.claude-fable-5). - **Account or region not enabled** — Bedrock requires the provider_data_share opt-in; Vertex AI requires signing the Advanced AI Safety Addendum and activating the model per region; Foundry requires Partner Model onboarding. - **Provider does not expose the model** — not all OpenAI-compatible gateways list Fable 5. Check the provider's current model catalog. - **Claude Code version too old** — Fable 5 requires Claude Code v2.1.170 or later. Older versions will not show Fable 5 in the model picker. - **Response is a refusal, not an error** — Fable 5 returns HTTP 200 with stop_reason=refusal. This is not an HTTP error code; it requires checking stop_reason in the response body, not monitoring HTTP status codes alone. - **Distinguishing Base URL, API key, model visibility, and provider-access problems** — Base URL determines the request destination; API key determines authentication; /v1/models determines model visibility; provider catalog determines whether the model is accessible on that platform. A failure at any layer will prevent a successful call.

AI API Doctor can help verify whether the API key, Base URL, and /v1/models response are configured correctly before sending a full request. Use it to narrow down which layer — URL, key, or model visibility — is causing the unavailability.

Setup or Check Steps

  • 1 Identify your platform: Claude API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, or OpenRouter
  • 2 Get the correct model ID for your platform (see provider table in the providers guide)
  • 3 Claude Code users: run claude update to upgrade to v2.1.170+, then select Fable 5 via /model or claude --model claude-fable-5
  • 4 Claude API / Claude Platform on AWS: configure beta fallbacks=[{model: claude-opus-4-8}] with header server-side-fallback-2026-06-01
  • 5 Bedrock / Vertex AI / Foundry: configure Anthropic SDK refusal-fallback middleware instead of fallbacks parameter
  • 6 Avoid deprecated parameters: temperature, top_p, top_k, and thinking=disabled are rejected with HTTP 400
  • 7 Call /v1/models to verify Fable 5 is visible on your account
  • 8 Run a small test request to confirm behavior

Common Errors

  • Using deprecated parameters (temperature/top_p/top_k/thinking=disabled) — returns HTTP 400
  • Monitoring only 4xx/5xx, missing HTTP 200 + stop_reason=refusal — refusals are invisible to error-based monitoring
  • Bedrock: using claude-fable-5 instead of anthropic.claude-fable-5 — model not found
  • Bedrock: not enabling provider_data_share opt-in — calls rejected silently
  • Vertex AI: not signing Advanced AI Safety Addendum — model not accessible
  • Claude Code: version < 2.1.170 — Fable 5 not in model picker
  • fallbacks beta header with wrong date — must be exactly 2026-06-01
  • Rate limit or overload on Fable 5 — fallbacks parameter does not cover these; they return as-is

Security / Billing / Permission Risks

  • Fable 5 safety classifier refusals return HTTP 200 with stop_reason=refusal — check stop_reason in your code, not just HTTP status codes
  • Beta fallbacks parameter is only available on Claude API and Claude Platform on AWS
  • Bedrock, Vertex AI, and Microsoft Foundry require Anthropic SDK refusal-fallback middleware — no server-side fallbacks
  • Claude Code requires v2.1.170+ to access Fable 5
  • Claude Opus 4.8 is the only permitted fallback target
  • Pricing and availability change — confirm against the Anthropic pricing page and your platform's model catalog before committing

When to Use AI API Doctor

After configuring Fable 5, use AI API Doctor to verify the API Key and Base URL are correct and that /v1/models returns claude-fable-5 in the model list.

When to Use LinkAI for Small Tests

If Anthropic Fable 5 pricing exceeds your budget, compare model options on LinkAI.

AI Summary

Claude Fable 5 access requires the correct per-platform model ID, refusal monitoring (stop_reason, not HTTP status), and per-platform fallback configuration — server-side beta fallbacks on Claude API/AWS, SDK middleware on Bedrock/Vertex/Foundry.

FAQ

What is the Claude Fable 5 model ID on each platform?
Claude API and Vertex AI: claude-fable-5 (lowercase with hyphens, a pinned snapshot). Amazon Bedrock: anthropic.claude-fable-5 (namespace prefix). OpenRouter: anthropic/claude-fable-5. The model ID format differs per provider — always confirm against the platform's model catalog before deploying.
How does Claude Fable 5 handle request refusals?
Fable 5 returns HTTP 200 with stop_reason=refusal when a safety classifier blocks a request. The stop_details.category field indicates the reason: cyber, bio, reasoning_extraction, or null. This is not an HTTP error — monitoring built around 4xx/5xx will never see refusals. You must check stop_reason in your response handler and configure a fallback to claude-opus-4-8 for blocked requests.
Does Claude Fable 5 work with Claude Code?
Yes, but Claude Code must be v2.1.170 or later. Older versions will not display Fable 5 in the model picker. Run claude update to upgrade, then select Claude Fable 5 via /model or use claude --model claude-fable-5 directly.
Can I use Claude Fable 5 via an OpenAI-compatible API?
The native Claude API, Bedrock, Vertex AI, and Foundry each use their own API (Messages API / Invoke / Converse API) — not OpenAI-compatible. If you need OpenAI compatibility, OpenRouter exposes Fable 5 at model ID anthropic/claude-fable-5 through its OpenAI-compatible endpoint. Other OpenAI-compatible gateways may also list Fable 5 — verify against their current model catalog.
Why is Claude Fable 5 unavailable in my account?
Fable 5 unavailability usually stems from one of these causes: the model is not in your /v1/models response (confirm against the provider catalog and account enablement status); the model ID format is wrong for your platform (Claude API and Vertex AI use claude-fable-5; Amazon Bedrock requires anthropic.claude-fable-5); your account or region has not been enabled yet (Bedrock requires provider_data_share opt-in; Vertex AI requires signing the Advanced AI Safety Addendum and activating per region; Foundry requires Partner Model onboarding); the provider does not currently expose the model for your plan or region; your Claude Code version predates Fable 5 support (requires v2.1.170+); or you are seeing a refusal rather than an error (Fable 5 returns HTTP 200 with stop_reason=refusal, not a 4xx code — check the response body, not just the HTTP status). To narrow down the cause, check each layer in order: Base URL determines the destination; API key determines authentication; /v1/models determines visibility; provider catalog determines platform-level availability. Run a /v1/models call to confirm whether the model is visible at all.

Not sure if your API works? Test before wiring it into production.

Check Base URL, API Key, model permissions, and usage signals first. Then use a small test budget to verify Claude Code, Cursor, or Cline before committing.