Autonomous multi-day coding sessions with self-verification

The model is designed for features that span an entire codebase, code review, and performance work. It can write its own tests to check its work and use vision to evaluate outputs against the original design or goal, reducing the need for human intervention.

Business outcome: Engineering teams can delegate complex, multi-day coding tasks and review completed work rather than supervising every step.

Document-heavy analysis with native diagram understanding

Claude Fable 5.1 understands diagrams, charts, and tables nested in files and PDFs. This capability improves document-heavy work in finance, legal, analytics, and architecture, allowing the model to extract and reason over complex visual data.

Business outcome: Reduces manual data extraction and review time in professional services that rely on dense, visual documents.

Built-in cybersecurity and biology risk classifiers

The model includes robust safeguards for cybersecurity and biology. Many queries in these domains are automatically routed to less capable models if flagged, with no charge for rerouted requests, helping organizations manage dual-use risk.

Business outcome: Provides a governance layer for sensitive domains, though it requires accepting a fallback model for flagged queries.

75% reduction in prompt cache read pricing

Cache reads now cost $0.25 per million tokens, 75% less than Fable 5. This reduces the cost of typical workloads by an estimated 25% and highly agentic workloads by up to approximately 45%.

Business outcome: Makes a frontier-class model financially viable for high-volume, agentic workloads that were previously cost-prohibitive.

Always-on reasoning for ambiguous tasks

Fable 5.1 runs adaptive thinking always on, aiming to avoid easy-seeming shortcuts and fix root causes of problems. This is designed to improve judgement on ambiguous work and reduce confident wrong answers.

Business outcome: Delivers more reliable outcomes on complex, open-ended problems where a wrong answer is costly.

Flexible enterprise data controls

Using Fable requires 30-day data retention for safety monitoring by default. Eligible enterprise customers can use Enterprise Frontier Safeguards to store data on their own cloud infrastructure, with human review done by the customer by default.

Business outcome: Offers compliance flexibility for enterprises with strict data residency and review requirements.

• Long-running agentic tasksBuilt for jobs that take hours and span many applications, operating unattended as a managed agent.
• Cache economics at scaleThe 75% cut in cache-read pricing makes frontier intelligence economical for high-volume, repetitive workloads.
• Root-cause problem solvingDesigned to fix root causes rather than symptoms, reducing the iteration cost on complex engineering problems.
• Enterprise data controlsOffers zero data retention and customer-managed review for eligible enterprises via Enterprise Frontier Safeguards.

• No reasoning-effort dialAdaptive thinking is always on; developers cannot turn down reasoning effort to save on token costs like they can with some competitors.
• Silent request routingSafety classifiers can silently route a request to a less capable model, which may surprise teams expecting Fable-level output on every query.
• Higher cost per taskIndependent analysis puts Fable 5.1 at $3.76 per Intelligence Index task versus $1.67 for GPT-6 Astra, because Astra spends fewer tokens.
• Benchmark caveatsScores on OSWorld 2.0 and AutomationBench were zero where safeguards intervened, and competitor scores are not shown due to differing task files.

Whole-codebase feature development

Engineering teams can hand off features that span an entire codebase, with the model writing its own tests to verify work and using vision to check outputs against the original design.

Unattended managed agents

Businesses can run Fable 5.1 as a managed agent on the Claude Platform to work through backlogs or pick up requests in Slack through Claude Tag, recovering from failed steps autonomously.

Scientific research agents

Research teams can deploy the model for complex analysis that demands sustained reasoning, with vendor claims noting research capabilities offering an early glimpse of AI contributing to scientific progress.

Document-heavy professional services

Finance, legal, and analytics teams can use the model's vision capabilities to understand diagrams, charts, and tables nested in files and PDFs, improving the speed of document review.

Identify a specific long-running, multi-step task in your workflow that currently requires heavy human supervision, such as a code review backlog or a deep research project.

Access Claude Fable 5.1 via the Claude Platform, the Anthropic API using the claude-fable-5-1 model identifier, or through AWS, Google Cloud, or Microsoft Foundry.

Review the safeguards and data retention policies, especially if you operate in cybersecurity or biology domains where requests may be routed to less capable models.

Run a pilot on a single high-value task, measuring the cost per completed task including cache reads, to validate the economic case before scaling to full agentic deployment.