Claude Fable 5

Supported Functionality

Item	Specification
Input	Text, images
Output	Text
Context	1,000,000 tokens
Max Output	128,000 tokens
Vision	✓ Supported
Function Calling	✓ Supported

Description

Claude Fable 5 is an Anthropic model released on June 9, 2026, with the official Claude API ID claude-fable-5. Built for demanding reasoning, software engineering, visual analysis, and long-horizon knowledge work, it accepts text and images, produces text, and provides a one-million-token context window with up to 128,000 output tokens. Anthropic currently marks it as active but legacy and recommends evaluating Fable 5.1 for new work.

Fable 5 shares an underlying model with the limited-access Mythos 5, but Fable adds safety classifiers that may route some sensitive requests to another Claude model. Anthropic reports meaningful gains over earlier public Claude models in complex coding, autonomous work, knowledge tasks, and vision, particularly for workflows that require planning, execution, and correction over time.

Key Capabilities

• Long-Horizon Autonomy: Plans, advances, and corrects a task across tool calls over extended work sessions.
• Large-Codebase Engineering: Understands cross-file dependencies for migration, debugging, refactoring, and review.
• Complex Reasoning and Research: Connects sources, calculations, and intermediate notes to solve multistep problems.
• Long Context and Durable Notes: Finds related information in extensive input and uses external notes to sustain task state.
• Visual Understanding: Extracts information from screenshots, charts, and other images for interface analysis and visually guided work.
• Tool and Agent Collaboration: Uses programmatic tools for retrieval, file operations, and validation.
• Safety Classifier Fallback: Applies additional checks to certain risk areas and may route a request to another model.

Technical Strengths

Feature	Benefit
Always-On Adaptive Thinking	Organizes reasoning according to task complexity, with depth controlled through effort.
One-Million-Token Context	Handles large collections of source material, code, and history without extensive manual splitting.
128K Output Limit	Allows substantial reports, code artifacts, and multipart analyses.
Long-Horizon Agent Stability	Integrates tool results and notes across sustained work, reducing goal loss and repeated exploration.
Joint Visual and Text Understanding	Combines evidence from screenshots and document images with written instructions.
Layered Safety System	Makes refusals identifiable so applications can handle them and choose suitable fallback behavior.

Frequently Asked Questions

Is Claude Fable 5 still available?

Yes. Anthropic marks it Active (legacy) and retains a distinct official model ID. New projects should also evaluate Fable 5.1, which Anthropic recommends for improved performance.

How can I use Fable 5's one-million-token context reliably?

Label sources, dates, and sections before posing explicit cross-checking questions; a large window does not guarantee every detail will be used correctly. For sustained tasks, keep key decisions and open items in an external record and require evidence locations in the response.

How can Fable 5's safety fallback affect an application?

Sensitive requests may trigger a classifier, causing this model to decline or route the request elsewhere. Check refusal markers as well as HTTP status, record the event, and decide whether to explain it to the user or retry according to your workflow.

How are Fable 5 and Mythos 5 related?

They share underlying capabilities, but Mythos 5 is restricted to approved organizations and uses a different safeguard arrangement in some areas. General applications should be built around Fable 5's public behavior, not assumed Mythos access.

When should I move from Fable 5 to Fable 5.1?

Prioritize testing 5.1 for long coding sessions, multistep research, document and spreadsheet work, or computer use. Recheck forced tool selection and thinking-block replay during migration because 5.1 changes those API behaviors.

What should I verify Fable 5 on Linkmodel?

Check the live LinkModel model page for its own ID, request format, context limits, and tool support. Test safety refusals, long-task timeouts, and multi-turn tool loops on a small scale before connecting an existing agent.