Claude Fable 5: Anthropic’s New Mythos-Class Model for Long-Horizon Workflows

Released on June 9, 2026, Claude Fable 5 represents a new paradigm in Anthropic’s AI lineup. Positioned as their first “Mythos-class” model, Fable 5 is specifically engineered to handle complex, long-horizon tasks—such as autonomous software engineering, multi-step scientific research, and advanced legal analysis—that require deep reasoning and extended planning.

Here is a comprehensive breakdown of Fable 5’s capabilities, architecture, novel safety features, and availability.

1. Technical Specifications & Pricing

Fable 5 is designed for high-performance agentic workflows. To support long-horizon tasks, Anthropic has given Fable 5 a massive context window and significantly increased output capabilities:

• Context Window: 1,000,000 tokens (with zero additional cost for long context usage).
• Max Output Limit: 128,000 tokens per single request.
• Knowledge Cutoff: January 2026.
• Pricing:
  • Input Tokens: $10.00 per million tokens.
  • Output Tokens: $50.00 per million tokens.
  • Prompt Caching: $12.50 per million tokens for cache writes, and $1.00 per million tokens for cache reads.

2. Benchmark Performance & Capabilities

Fable 5 establishes a new state-of-the-art for frontier large language models. It is built to excel in autonomous agentic loops where models must make plans, use tools, and self-correct over hundreds of steps.

Agentic Autonomy

Fable 5’s main selling point is its stability inside agentic loops (like Claude Code or local developer setups). It features highly refined tool-use mechanics, leading to fewer hallucinations during execution and a higher success rate in executing terminal commands and correcting its own syntax errors.

3. The Safety Architecture: Session Routing & Fallbacks

Fable 5 introduces a new built-in safety classifier system designed to prevent misuse in high-risk categories (such as cyberweapons, chemistry, and biological agents):

• The Routing Mechanism: If Fable 5’s real-time safety classifiers detect that a prompt falls into a restricted or dangerous domain, the system automatically and silently routes the query to a secondary, more restricted model (Claude Opus 4.8).
• Trigger Rate: Initial data shows this fallback routing triggers in approximately 5% to 9% of user sessions.
• Data Retention Policy: To continually train and audit these safety classifiers, Anthropic retains prompts and outputs for Fable 5 sessions for up to 30 days. This represents a temporary departure from the Zero Data Retention (ZDR) policy standard on other Anthropic enterprise endpoints.

4. Claude Mythos 5: The Restricted Sibling

Alongside the public launch of Fable 5, Anthropic introduced Claude Mythos 5.

Mythos 5 is the exact same underlying model as Fable 5, but with the safety classifiers and routing filters completely lifted. Because of the inherent risk of an unfiltered Mythos-class model, Mythos 5 is not publicly available via API or web interface.

Instead, access is restricted to a small, strictly vetted group of research institutions and defense organizations—specifically those collaborating on Project Glasswing—for cybersecurity defense, national infrastructure auditing, and advanced safety alignment research.

5. Availability

Claude Fable 5 is available starting June 9, 2026, across the following channels:

• Web Interfaces: Claude.ai chat interface and Claude Code CLI.
• Developer API: Available directly on the Anthropic Console.
• Cloud Partners: Accessible via AWS Bedrock, Google Cloud (Vertex AI), and Microsoft Foundry.
• IDEs: Integrated into major coding assistants, including GitHub Copilot.

6. How to Use Claude Fable 5

Integrating Fable 5 into your developer workflows is simple using the official Anthropic SDKs. Given the model’s massive 1M context window, leveraging prompt caching is highly recommended to control token costs and query latency.

Node.js SDK Example

import Anthropic from '@anthropic-ai/sdk';

const anthropic = new Anthropic({
  apiKey: process.env.ANTHROPIC_API_KEY,
});

const response = await anthropic.messages.create({
  model: 'claude-5-fable-20260609',
  max_tokens: 4096,
  system: [
    {
      type: 'text',
      text: 'You are an autonomous senior developer agent. Analyze the provided repository structure and plan the upgrade.',
      cache_control: { type: 'ephemeral' } // Active Cache
    }
  ],
  messages: [
    { role: 'user', content: 'Design a system migration blueprint from Postgres to Supabase...' }
  ],
});

console.log(response.content[0].text);

Python SDK Example

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-5-fable-20260609",
    max_tokens=4096,
    system=[
        {
            "type": "text",
            "text": "You are a legal research bot auditing security compliance clauses.",
            "cache_control": {"type": "ephemeral"}
        }
    ],
    messages=[
        {"role": "user", "content": "Audit this cloud hosting contract for compliance..."}
    ]
)

print(message.content[0].text)