Anthropic launches Claude Fable 5 with guarded fallback

Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026, per the Anthropic announcement and API documentation. Fable 5 is described as a Mythos-class model made safe for general use, with capabilities that exceed any model Anthropic has previously made generally available. Claude Mythos 5 is the same underlying model with safety classifiers lifted in some areas, initially deployed through Project Glasswing in collaboration with the US government as an upgrade to Claude Mythos Preview. Both models are available at $10 per million input tokens and $50 per million output tokens, double the price of Opus 4.8, with batch API pricing at $5/$25.

Anthropic's API documentation lists support for a 1M-token input context window and up to 128k output tokens for both models, per the API docs. Vision, tool use, memory, compaction, and adaptive thinking are supported. The model ID is claude-fable-5; both models are available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry from June 9.

When Fable 5's classifiers decline a request, the API returns a normal HTTP 200 with stop_reason: "refusal" and a stop_details object whose category field names the triggered classifier - "cyber", "bio", or "reasoning_extraction" - per Anthropic's API documentation. Refusals before any output leave content empty and are not billed. A beta fallbacks parameter lets integrators specify up to three fallback models (tried in order) in a single API call; the request must include the server-side-fallback-2026-06-01 beta header. Server-side fallback is available on the Claude API and Claude Platform on AWS only, and is explicitly not available on the Message Batches API, Amazon Bedrock, Vertex AI, or Microsoft Foundry; the SDK middleware (TypeScript, Python, Go, Java, C#) provides a cross-platform alternative that sends the fallback-credit beta header automatically.

Fable 5 runs classifiers on three areas, per Anthropic's announcement: cybersecurity (exploit and offensive cyber tasks), biology and chemistry (dual-use life sciences requests), and distillation (attempts to extract capabilities for training competing models). When any classifier fires, the request falls back to Opus 4.8; users are informed whenever this occurs. Anthropic states that in more than 95% of Fable 5 sessions no fallback occurs at all. External red-teaming found no universal jailbreaks in over 1,000 hours of bug bounty testing, though the UK AISI made progress toward one in a brief initial window.

On the Claude API and consumption-based Enterprise plans, Fable 5 is fully available from June 9, per Anthropic. On subscription plans (Pro, Max, Team, seat-based Enterprise), Fable 5 is included at no extra cost through June 22; after that date, usage credits are required until Anthropic restores it as a standard plan feature. Mythos 5 is restricted to Glasswing cybersecurity partners (cyber safeguards lifted) and, soon, select biology researchers (biology and chemistry safeguards lifted with cyber safeguards still active).

• •Fallback beta maturity: the server-side-fallback-2026-06-01 beta is explicitly versioned; teams building on it should track when it moves to stable and update the header and response shapes.
• •Trusted access expansion: Anthropic plans to add biology researchers to the Mythos 5 trusted access program and to expand Glasswing partnerships, per the announcement.
• •Subscription capacity: Anthropic states it will restore Fable 5 as a standard plan feature when capacity allows but has not given a date.