This guide walks you through accessing Claude Fable 5, configuring it for different workflows, and making the most of its unique capabilities — from autonomous agents to vision-based self-validation.

Accessing Claude Fable 5

Claude Fable 5 is available across multiple surfaces depending on your subscription:

• Claude Web, Desktop, and Mobile: Select “Fable 5” from the model picker in the top-left corner of the interface.
• Claude Code: Requires version 2.1.170 or later. Update your CLI with claude update, then select Fable 5 when starting a session.
• Claude API: Use the model ID claude-fable-5 in your API calls. Pricing is $10 per million input tokens and $50 per million output tokens, with a 90% discount on prompt-cached input tokens.

Fable 5 is included for Pro, Max, Team, and Enterprise users. During the promotional period (July 1–19, 2026), you can use up to 50% of your weekly subscription limits on Fable 5 at no extra cost. After that threshold, you can continue with usage credits or switch to another model.

Using Fable 5 for Autonomous Coding

The most transformative use of Claude Fable 5 is autonomous, multi-day coding. Unlike previous models that lose context or produce inconsistent results over long sessions, Fable 5 sustains quality across days of unattended work.

Step 1: Define the task clearly. Fable 5 performs best when you provide a well-structured brief — the scope, constraints, testing requirements, and success criteria. For example: “Migrate all Ruby controllers from Rails 6 to Rails 7 syntax across the api/ directory. Write tests for each migrated file. Report any files where the migration fails.”

Step 2: Let it run in an agent harness. Use Claude Code or Claude Managed Agents as the execution environment. Fable 5 will plan across stages, delegate sub-tasks, write its own tests, and validate results against your original goals — all without step-by-step human instruction.

Step 3: Review completed work, not intermediate steps. The paradigm shifts from supervising every action to reviewing finished deliverables. Check the final output, test results, and any flagged issues. Fable 5 catches most problems during its own self-validation loop, so your review focuses on strategic decisions rather than code-level bugs.

Leveraging Vision and Document Intelligence

Fable 5 understands diagrams, charts, and nested tables inside PDFs and files — a capability that transforms document-heavy workflows in finance, legal, and analytics.

Upload a financial report with embedded charts and ask: “Extract all revenue figures from the bar charts on pages 12–15 and reconcile them with the table on page 3.” Fable 5 reads both the visual and tabular data, cross-references them, and flags discrepancies — work that previously required manual extraction and reconciliation.

For legal teams, Fable 5 can review redlined contracts, compare clause variations across multiple document versions, and produce consolidated markups that lawyers have rated as matching or beating prior models in blind review.

Understanding the Safety Fallback System

Fable 5 includes robust safeguards for cybersecurity and biology domains. When a query is flagged by these classifiers — which trigger in less than 5% of sessions on average — the model automatically routes the response to Claude Opus 4.8 instead. You are not charged Fable 5 pricing for rerouted requests.

This means most users will never notice the fallback. If you do experience a model switch mid-conversation, it is the safeguard system working as designed. For legitimate cybersecurity and biology research, Anthropic offers access to Claude Mythos 5 through trusted access programs — sign up on the Mythos access interest form on Anthropic’s website.

Optimizing Token Usage and Cost

Fable 5 is more token-efficient than previous Claude models. On FrontierBench evaluations, it achieves top scores even at medium effort levels, meaning you often get frontier-quality results without burning maximum reasoning tokens.

To minimize cost:

• Use prompt caching: Cache your system prompts, project context, and recurring instructions to receive a 90% input token discount.
• Set effort levels appropriately: Not every task needs maximum reasoning. Use lower effort for straightforward tasks and reserve high effort for complex, multi-stage work.
• Batch related queries: Fable 5 maintains context across long conversations, so grouping related tasks in one session reduces redundant context loading.

When to Choose Fable 5 vs. Other Claude Models

Fable 5 is not the right model for every task. Use it when the work is genuinely ambitious — large migrations, multi-day agent sessions, deep research, or document analysis that requires vision. For quick questions, short-form writing, or routine chat, Claude Sonnet 4 or Opus 4.8 offer faster responses at lower cost.

The rule of thumb: if a task would take you all afternoon and previous AI models couldn’t complete it reliably, hand it to Fable 5. If it’s a 30-second question, use a faster model.