Claude Fable 5.1

Supported Functionality

Item	Specification
Input	Text, images
Output	Text
Context	1,000,000 tokens
Max Output	128,000 tokens
Vision	✓ Supported
Function Calling	✓ Supported; forced tool choice is not supported

Description

Claude Fable 5.1 is an Anthropic language model released on September 1, 2026, with the official Claude API ID claude-fable-5-1. It targets demanding reasoning, long-running agent tasks, and complex knowledge work, accepts text and images, and offers a one-million-token context window with up to 128,000 output tokens.

Compared with Fable 5, it improves agentic coding across hours, multistep research, document and spreadsheet production, dense-chart understanding, and computer use. Adaptive thinking is always on, with depth controlled through effort; safety classifiers may decline certain sensitive requests and route them to another Claude model.

Key Capabilities

• Long-Horizon Agentic Coding: Works through multi-file features, refactors, migrations, debugging, and reviews while retaining the goal across tool calls.
• Multistep Research: Follows up on retrieved findings, checks evidence, and synthesizes structured conclusions.
• Complex Knowledge Work: Carries analysis through to documents, formula-based spreadsheets, and slide decks.
• Long-Context Reasoning: Connects details distributed across extensive documents, code, and conversation history.
• Visual Material Understanding: Reads dense charts, filings, and nested PDF tables, using crop-and-zoom tools for closer inspection.
• Computer Use: Performs multistep tasks in browsers and desktop applications and recovers from failed actions.
• Tool Collaboration and Safety Fallback: Supports automatic and strict tool use, but not forced any or named-tool selection.

Technical Strengths

Feature	Benefit
Always-On Adaptive Thinking	Adjusts reasoning to the task without requiring a fixed thinking budget.
Adjustable effort	Varies thinking depth between difficult decisions and routine steps in a long workflow.
Large Context and Output	Holds substantial code, evidence, and operation history while allowing complete deliverables.
Long-Horizon Tool Execution	Maintains objectives and progress across extended coding, research, and artifact workflows.
Structured Tool Use	Combines automatic tool selection with strict schemas and structured outputs for verifiable agents.
Safety Classification and Fallback	Exposes refusal signals so applications can record the reason and choose a suitable alternate path.

Frequently Asked Questions

How can I keep a long Claude Fable 5.1 task on track?

Provide a concrete goal, acceptance criteria, and a durable progress record, saving intermediate work to files or tools. Avoid rewriting early conversation turns: Anthropic says its thinking blocks are bound to the preceding history and may become invalid after such edits.

Why does a Claude Fable 5.1 tool request return a validation error?

Using tool_choice with type: "any" or a forced named tool returns an error. Use the default auto mode, apply strict: true or structured outputs for schema validity, and describe in the prompt when the tool should be used.

What should I test when migrating from Fable 5 to Fable 5.1?

Beyond changing the official model ID, test forced-tool settings, thinking-block replay, and any code that edits conversation history. Older Claude models cannot read thinking blocks produced by Fable 5.1, so model routing and fallback need dedicated tests.

Should I fill the entire one-million-token window with documents?

No. Group sources by topic, label provenance and versions, and supply retrieval cues for important questions before asking for cross-source analysis; then check omissions, stale material, and conflicts in the answer.

When should I select Fable 5.1 instead of Opus 5?

Use Fable 5.1 when Opus 5 at higher effort still misses your acceptance criteria on long-running coding, deep research, or complex knowledge work. Anthropic recommends starting most workloads with Opus 5 and choosing from task-specific evaluations.

How should I validate Claude Fable 5.1 on LinkModel?

Check LinkModel's live model page for its public ID, request schema, image input, and tool support, then run small tests covering long context, tools, and refusal handling. Do not assume Anthropic-specific beta headers or thinking-block options pass through unless LinkModel documents them.