Claude Fable 5 explained: capabilities, pricing, availability, safeguards, and best use cases

01

What Claude Fable 5 is

Claude Fable 5 is Anthropic's fifth-generation frontier model for demanding coding, agentic execution, vision, and professional knowledge work. Anthropic released it in June 2026 and restored global access on July 1 after a temporary suspension connected to US export controls.

Fable is not intended to replace every Claude model. Anthropic positions it above everyday workhorse models for projects that are difficult, long-running, asynchronous, or expensive to supervise turn by turn. The API model identifier is claude-fable-5.

02

Long-horizon agents and coding

Anthropic describes Fable 5 as capable of working for days in an agent harness such as Claude Code or Claude Managed Agents. The model can plan across stages, delegate work, inspect results, write tests, and use vision to compare an implementation with a design goal.

These capabilities depend on the surrounding system. A production agent still needs scoped tools, repository permissions, checkpoints, test commands, cost and time budgets, observability, and human approval for consequential changes. Model autonomy should increase only after the workflow proves reliable.

• Large migrations and complex implementations
• Multi-stage research and document-heavy analysis
• Asynchronous agent work with review checkpoints
• Vision-assisted coding and interface evaluation

03

Knowledge work and vision

Fable 5 is designed to interpret diagrams, charts, tables, files, and PDFs as part of larger analytical tasks. This can support finance, legal review, analytics, architecture, research, and planning, provided that the user supplies authoritative material and verifies important conclusions.

For evaluation, use messy real cases rather than polished demos: conflicting documents, scanned tables, missing evidence, ambiguous requirements, and long chains of calculations. Score claim traceability, extraction accuracy, uncertainty, and the usefulness of the final deliverable.

04

Pricing, prompt caching, and data retention

Anthropic lists Fable 5 at $10 per million input tokens and $50 per million output tokens. Prompt caching retains the existing 90% input-token discount for qualifying cached input. US-only inference is listed at 1.1 times the standard token price.

Fable use requires 30-day data retention for safety monitoring according to Anthropic's product page. Teams should verify current terms, regional availability, and organizational requirements before sending confidential or regulated information.

05

Safeguards and fallback behavior

Fable 5 includes additional safeguards for high-risk cybersecurity and biological requests. Anthropic states that many flagged requests are routed to Claude Opus 4.8 instead, with API customers using the provider's fallback configuration.

This matters in ordinary evaluation because conservative classifiers can also affect legitimate coding or research tasks. Record refusals, fallbacks, false positives, and successful completions rather than treating every response as if it came from the requested model.

06

When Fable 5 is worth using

Use Fable when a difficult task benefits from fewer interventions, sustained reasoning, strong coding, complex files, or a long agent trajectory. Do not use it automatically for classification, formatting, simple extraction, short support answers, or high-volume routine work.

A practical architecture lets a smaller model handle routine execution while Fable creates the plan, reviews difficult cases, or validates the final result. Anthropic itself presents this advisor-style orchestration pattern as a way to preserve frontier reasoning while controlling token cost.