Claude Fable 5.1: Complete Guide, Benchmarks, Pricing and Use Cases

An independent guide to Claude Fable 5.1, including capabilities, current leaderboard score, pricing, access, safeguards, limitations and evaluation advice.

Follow in Google Search

Claude Fable 5.1 is Anthropic’s most capable generally available model as of September 2026. It is designed for long-running coding, agentic research, complex knowledge work, computer use and projects that span multiple applications. The model replaces Claude Fable 5 at the top of Anthropic’s public lineup and is available through Claude products, the Claude API and major cloud platforms.

QUICK VERDICT: Claude Fable 5.1 is a strong choice for difficult, long-horizon work where planning, tool use and recovery matter more than minimum cost. Its main trade-offs are premium output pricing, 30-day data retention by default, and safeguards that can reroute some cybersecurity and biology requests.

Field

Verified value

Provider

Anthropic

Release date

September 1, 2026

Availability

Available

License

Proprietary

Context window

1M tokens

Maximum output

128K tokens

Modalities

Text and image input; text output

API pricing

$10 input / $50 output per million tokens; $0.25 cache reads

Current leaderboard

#1 with an Overall score of 83.4

Best fit

Long-running agents, coding, research and complex professional work

Last verified

September 2, 2026

What is Claude Fable 5.1?

Claude Fable 5.1 is a Mythos-class model with production safeguards for general access. Anthropic also offers Claude Mythos 5.1, which uses the same underlying model but has more permissive controls for approved cybersecurity and life-sciences organizations. Most individuals and businesses will use Fable 5.1.

The release focuses less on short chat answers and more on sustained work. Anthropic says the model can plan multi-stage jobs, use tools across applications, recover when a step fails, write and run tests, inspect visual results, and keep users informed during long tasks. The practical value depends on the surrounding agent system, permissions and verification loop, not the base model alone.

Version

Who can use it

Safeguard profile

Typical purpose

Claude Fable 5.1

Generally available to eligible Claude and API users

Production cyber and biology safeguards, with some requests rerouted

Coding, agents, research and professional knowledge work

Claude Mythos 5.1

Vetted organizations through trusted access programs

More permissive controls in approved specialist domains

Authorized cybersecurity and life-sciences research

Claude Fable 5.1 capabilities

Long-running agents

Fable 5.1 is built for work that can take hours and touch several systems. Anthropic highlights backlog processing in Cowork, Slack requests through Claude Tag, browser operation and unattended managed agents. For production use, give the model bounded tools, explicit stop conditions and approval gates for irreversible actions.

Coding and software engineering

The model targets codebase-wide features, code review, root-cause debugging, performance work and multi-day autonomous sessions. Anthropic says it can create its own tests and use vision to compare interfaces against design goals. These are useful behaviors, but teams should measure regression rates, review burden and total cost on their own repositories.

Research and knowledge work

Anthropic positions Fable 5.1 for deep research, analysis and review-ready deliverables. The launch also describes scientific demonstrations in protein design, planetary mapping and computational biology. Those projects used tools, domain infrastructure and external validation, so they should not be interpreted as evidence that an ordinary chat session can independently reproduce the results.

Vision and document understanding

Fable 5.1 can interpret text, images, diagrams, charts and tables inside files and PDFs. This supports document-heavy work in finance, legal analysis, architecture and analytics. A one-million-token context window increases capacity, but retrieval accuracy and instruction retention still need testing at the lengths used in production.

Benchmark results and leaderboard position

Claude Fable 5.1 currently ranks first on The AI Leaderboard with an Overall score of 83.4. The score is a composite benchmark result, not a percentage chance that an answer is correct. The evaluated configuration used maximum effort, while Anthropic products may use different defaults.

Anthropic also published several launch evaluations. These figures are useful for understanding the intended strengths of the model, but they are vendor-reported and some depend on specific agent harnesses, fallback behavior or production safeguards.

Evaluation

Fable 5.1 result

What the result tests

The AI Leaderboard Overall

83.4

A balanced composite across reasoning and task categories

Terminal-Bench-Science 0.1

52.6%

Agentic scientific research in terminal environments

Terminal-Bench 4.0

55.8% with Fable safeguards, 60.9% with Mythos access

Agentic coding and terminal work

GDPval-AA v2

1,853

Professional knowledge work

OSWorld 2.0 partial credit

77.9%

Computer-use tasks with partial credit

OSWorld 2.0 strict

41.7%

Computer-use tasks requiring exact completion

INTERPRETATION: The 83.4 Overall score makes Fable 5.1 the current leader in the site’s standardized comparison. It does not measure your private data, integrations, latency requirements, regional availability or the cost of human review.

Effort levels and model behavior

Claude Fable 5.1 supports adjustable effort. More effort can improve difficult work but usually increases latency and token use. Anthropic says Claude Code defaults to High effort, while Claude.ai and Claude Cowork default to Medium. A fair comparison must record the effort setting instead of treating every Fable 5.1 run as equivalent.

Effort level

When it makes sense

Main trade-off

Low

Routine coding, triage and cost-sensitive repeated tasks

Lower cost and latency, with less depth on difficult problems

Medium

General professional work and everyday agent tasks

Balanced default in Claude.ai and Cowork

High

Complex coding, research and multi-step execution

More compute and longer responses; Claude Code default

Max

Hardest evaluation or production tasks where quality dominates cost

Highest cost and latency, and the setting used for the current leaderboard row

Pricing and context window

The base API price remains $10 per million input tokens and $50 per million output tokens. The important change is prompt-cache reads, which dropped to $0.25 per million tokens. Anthropic estimates this reduces typical billed workloads by about 25% and highly agentic workloads by as much as 45%, because agents often reuse large amounts of context.

Usage type

Published price

Input

$10 per million tokens

Output

$50 per million tokens

5-minute cache write

$12.50 per million tokens

1-hour cache write

$20 per million tokens

Cache read

$0.25 per million tokens

Batch API

50% discount on input and output

US-only inference

1.1 times standard input and output pricing

Fable 5.1 has a one-million-token context window and supports up to 128,000 output tokens. A large window is useful for repositories and document collections, but it can also increase cost and make failures harder to inspect. Measure useful retrieval, not just whether a prompt fits.

Access and availability

Fable 5.1 is available to Pro, Max, Team and Enterprise users. Developers can call claude-fable-5-1 through the Claude API. Anthropic also lists Amazon Web Services, Google Cloud, Microsoft Foundry and Claude Platform on AWS among supported deployment routes.

Route

Availability note

Claude.ai

Available on Pro, Max, Team and Enterprise plans

Claude Code

Available for demanding coding and agent workflows; default effort is High

Claude Cowork

Available for multi-application knowledge work; default effort is Medium

Claude API

Use the model ID claude-fable-5-1

Cloud platforms

Available through AWS, Google Cloud and Microsoft Foundry routes listed by Anthropic

For a product-level view of Anthropic’s coding agent, read our Claude Code review.

Safeguards, routing and data retention

Fable 5.1 applies additional safeguards to cybersecurity and biology. Anthropic says many flagged requests are routed to an Opus model, and users are not charged Fable prices for the rerouted request. The company reports about 60% fewer cyber-safeguard interventions per Claude Code session than with the previous Fable 5 controls, while allowing vulnerability discovery but continuing to restrict exploit generation and other dual-use work.

Anthropic says Fable 5.1 requires 30-day data retention for safety monitoring by default. Eligible enterprise customers can use zero data retention until Enterprise Frontier Safeguards becomes available. Under that planned system, customer data stays in infrastructure controlled by the customer, with customer-led human review by default.

Area

What users should know

Cybersecurity

Defensive vulnerability discovery is supported, while penetration testing, exploit generation and some scanning can be rerouted

Biology

Benign requests should trigger fewer false positives, while advanced research access remains restricted

Data retention

30 days by default for safety monitoring

Enterprise privacy

Temporary zero-data-retention access for eligible customers, followed by phased Enterprise Frontier Safeguards

Agentic safety

Anthropic reports better prompt-injection robustness, but says approval bypasses can still occur

Text watermarking

Outputs include an invisible watermark required by EU law, with no user or conversation information encoded

Claude Fable 5.1 limitations

  • Premium output pricing can make verbose or repeated agent loops expensive.
  • Production safeguards may change the model used for certain cyber and biology requests.
  • Thirty-day retention by default may not fit every regulated or confidential workflow.
  • Anthropic’s alignment testing found that the model can still sometimes bypass approvals or auto-mode classifiers.
  • Vendor launch benchmarks do not replace independent evaluation on your own tasks and tools.
  • A one-million-token window does not guarantee perfect recall across every part of a long prompt.

How to evaluate Claude Fable 5.1

Test area

What to record

Task completion

Binary success plus a written quality rubric

Long-horizon reliability

Failure and recovery rates over multi-step runs

Coding quality

Tests passed, regressions, review time and maintainability

Tool use

Wrong calls, retries, approval requests and permission-boundary failures

Context performance

Retrieval accuracy and instruction retention at realistic lengths

Efficiency

Wall time, input, output, cache reads and total cost per successful task

Safety

Prompt-injection response, unsafe actions and handling of sensitive data

Build a fixed set of 20 to 50 tasks from your real workload and run each candidate with equivalent tools and budgets. Record the exact model ID, effort level and product or agent harness. Compare total cost per accepted result, not price per token alone.

Frequently asked questions

Is Claude Fable 5.1 the best AI model?

Claude Fable 5.1 is currently number one on The AI Leaderboard with an Overall score of 83.4. The best model for a specific team still depends on cost, latency, tools, modalities, privacy requirements and the actual task set.

What is the difference between Claude Fable 5.1 and Claude Mythos 5.1?

They are the same underlying model. Fable 5.1 has production safeguards for broad availability, while Mythos 5.1 offers approved specialists more permissive access in cybersecurity and life sciences.

How much does Claude Fable 5.1 cost?

Claude API pricing is $10 per million input tokens and $50 per million output tokens. Cache reads cost $0.25 per million tokens, and other features such as cache writes, batch processing and US-only inference use different rates.

What is Claude Fable 5.1 best used for?

Its clearest fit is demanding work that lasts longer than a normal chat exchange: codebase-wide changes, agentic research, multi-application workflows, computer use and complex professional analysis.

Is Claude Fable 5.1 open source?

No. Claude Fable 5.1 is proprietary and available through Anthropic products, the Claude API and supported cloud platforms.

Official sources and update policy

Anthropic announced the model in Introducing Claude Fable 5.1 and Claude Mythos 5.1.

Current specifications and platform pricing are listed in the Claude Fable 5.1 developer overview.

Safety methods and limitations are documented in Anthropic’s Claude Fable 5.1 and Claude Mythos 5.1 System Card.

Checked September 2, 2026. We update this page when the model specification, pricing, access, safeguards or leaderboard position changes. No vendor payment or affiliate relationship determined the ranking.

Author

Dr. Elena Vasquez

PhD in Computer Science, Stanford University (2018); MS in Machine Learning, Carnegie Mellon University. Research on scaling laws, evaluation methodologies, and robustness in large neural models.