In a stunning move that effectively shifts the frontier of AI capabilities, Anthropic has officially announced the launch of Claude Fable 5.

The release marks a historic first: Anthropic is opening up public access to its top-secret, next-generation “Mythos-class” architecture—a hyper-capable tier that sits comfortably above its current flagship Opus models.

Historically, Anthropic heavily restricted this tier out of strict safety concerns, limiting it to a handful of cyber-defense agencies and elite biological research labs under Project Glasswing. With the deployment of a highly sophisticated, real-time automated safety routing system, Anthropic is finally making this massive computational powerhouse available to the public.

1. Defining the “Mythos” Class: Built for Long-Horizon Agency

If Haiku is built for speed, Sonnet for balance, and Opus for hard logic, the Mythos-class is engineered for prolonged, multi-day, asynchronous execution.

Instead of just responding to a prompt, Fable 5 is designed to function as an independent digital agent. It handles highly complex, open-ended tasks that would typically cause previous models to lose context or halt completely after a few turns.

• Autonomous Self-Correction: Fable 5 can continuously execute complex tasks for hours or days without human intervention. It builds its own evaluation harnesses, tests its output, and systematically updates its own internal technical skills based on what it learns during runtime.
• Massive Codebase Engineering: The model’s reasoning capabilities scale aggressively with task complexity. During early corporate beta testing, financial software giant Stripe utilized Fable 5 to execute a sprawling, system-wide codebase migration across a 50-million-line Ruby repository. A task that was projected to absorb an entire engineering squad for more than two months was successfully wrapped up by the model in a single day.
• Flawless Design Implementation: Backed by an advanced, next-generation vision engine, Fable 5 can parse complex engineering blueprints, financial charts, and nested multi-page PDFs. In coding environments, it uses its vision to visually critique its own web designs against original UI goals, fixing code alignment errors invisibly before final delivery.

2. The Twin Strategy: Fable 5 vs. Mythos 5

To balance commercial pressure against international safety standards, Anthropic is bifurcating the rollout into two models that share the exact same underlying neural weights but utilize entirely different safety frameworks:

Claude Fable 5 (The Safe-for-Everyone Twin)

Available for general corporate and developer use. It includes powerful, highly optimized safety classifiers that intercept queries touching on hazardous domains.

Claude Mythos 5 (The Restricted Unchained Model)

Operates with raw capabilities and zero defensive boundaries. This model is strictly firewalled and limited to vetted cybersecurity defenders, infrastructure providers, and life sciences research partners working directly alongside federal governments on biodefense screening and vulnerability discovery.

3. The Dynamic Airbag: Real-Time Safety Fallbacks

To ensure that Fable 5 remains safe for the public without suffering from annoying, rigid text refusals, Anthropic engineered a clever, seamless routing system.

[User Input Query] ──► Real-Time Core Classifier Scans for Hazardous Material
                                   │
                    ┌──────────────┴──────────────┐
                    ▼                             ▼
            [Trips Safety Filter]         [Passes Safety Filter]
            (Cyber / Bio / Chem)          (General / Code / Research)
                    │                             │
                    ▼                             ▼
        Reroutes to Claude Opus 4.8          Executes on Claude Fable 5
          (Charged at Opus rates)            (Charged at Premium Fable rates)

If a user inputs a query that trips safety thresholds for high-risk domains like biochemistry, cyber-warfare, or strict model distillation (cloning the model’s intelligence), the platform instantly and invisibly hands the prompt off to Claude Opus 4.8. The user receives their answer safely, and their token billing dynamically adjusts down to standard Opus pricing.

4. API Parameters and Premium Token Economics

As a premium tier model commanding immense datacenter compute, Fable 5 is priced at a exact premium—roughly double the cost of Claude Opus 4.8:

Access and Availability

The rollout is officially live across major cloud networks and native platforms:

• The API & Console: Accessible via the developer console using the string claude-fable-5.
• Cloud Infrastructure: Globally available on Amazon Bedrock (initially launching in US East and Europe Stockholm regions), Google Cloud Vertex AI, and Microsoft Foundry.
• Consumer Subscriptions: Pro, Max, Team, and Enterprise tier subscribers on Claude.ai get free, uninhibited access to Fable 5 from June 9 through June 22, 2026. Starting June 23, due to severe initial server capacity limits, Fable 5 usage will count as double against standard consumption limits or require usage credits.