Long-running agents
Use Fable for multi-stage projects that need sustained planning, tools, memory, and context management over hours or days.
Independent tool overview
Claude Fable at a glance
Claude Fable 5 is Anthropic's most capable generally available model, built for difficult software, research, and long-running agent work that can justify premium pricing and slower responses.
Visit the official Claude Fable site ↗
- Best for
- The hardest long-running agent, coding, research, and knowledge-work tasks
- Context window
- 1 million tokens
- Maximum output
- 128,000 tokens
- API model ID
- claude-fable-5
- Consumer access
- Paid Claude plans; not included on Free
- API price
- $10 input / $50 output per million tokens
Overview
What Claude Fable is
Claude Fable 5 sits above Claude Opus and Sonnet in Anthropic's model lineup. It is designed for the hardest long-horizon work: large software projects, multi-step research, complex knowledge work, and agents that operate for hours or longer.
The model accepts text and images, provides a 1 million-token context window, and can produce up to 128,000 output tokens. Its agent stack includes memory, code execution, programmatic tool use, context editing, and compaction for work that cannot fit cleanly into one prompt-response cycle.
Fable also has stricter safety classifiers than Anthropic's other generally available models. Applications must treat a refusal as a normal successful API response and decide whether to retry, route to another model, or ask the user to revise the request.
Use cases
Who Claude Fable is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
High-stakes software work
It is suited to difficult repository-scale implementation, debugging, architecture, and technical investigation where model quality matters more than speed.
Complex research and analysis
The large context window and tool support make it useful for synthesizing extensive documents, data, images, and external evidence.
Large multimodal projects
Fable can reason across text and images when a project includes diagrams, screenshots, scanned documents, or other visual inputs.
Capabilities
Core Claude Fable features
Always-on adaptive thinking
Fable automatically determines how much internal reasoning a task needs. Developers can tune effort, but cannot disable thinking entirely.
One-million-token context
The model can work across very large codebases and document collections, with compaction and context-editing options for longer-running sessions.
Agent tools
Anthropic supports memory, code execution, programmatic tool calling, tool-result clearing, task budgets, and other controls for durable agent workflows.
Text and vision input
Fable accepts text and images and can produce long text responses, making it useful for mixed document and visual analysis.
Prompt caching
Cached prompt reads cost 90% less than standard input tokens, which can materially reduce repeated-context costs in large applications.
Safety-aware fallback handling
A refusal arrives with HTTP 200 and a refusal stop reason, allowing applications to route safely to a fallback or request a revised prompt.
Process
How the Claude Fable workflow works
Step 1
Reserve Fable for the hardest work
Route routine requests to a faster, less expensive model and use Fable when evaluations show a meaningful quality advantage.
Step 2
Set effort and task budgets
Give the model clear success criteria, tool permissions, and resource limits for long-running work.
Step 3
Manage context deliberately
Use prompt caching for stable instructions and references, then use compaction or context editing as the session grows.
Step 4
Handle refusals as an application state
Inspect the stop reason even when the request succeeds, and define a safe fallback or user-revision path.
Step 5
Evaluate quality, latency, and cost
Test representative tasks against Opus and Sonnet before making Fable the default model.
Cost
Claude Fable pricing and free plan
Fable is Anthropic's premium model. API use costs $10 per million input tokens and $50 per million output tokens. Paid Claude-plan access varies: some plans use credits from the first message, while eligible premium plans include a limited weekly allowance before credits apply.
Claude Pro
$20/month or $200/year
Fable is available through usage credits rather than the standard included Pro allowance.
- No Fable access on the Free plan
- Annual billing works out to about $17 per month
- Available across supported Claude apps
Claude Max
From $100/month
Max plans include Fable use for up to 50% of the weekly usage limit, after which usage credits apply.
- Max 5x costs $100 per month
- Max 20x costs $200 per month
- Actual included use depends on token consumption and workload
Claude Team
From $20/seat/month annually
Standard Team access uses credits from the start; eligible premium seats include Fable for up to 50% of the weekly allowance.
- Standard seats are $20 annually or $25 monthly
- Premium seats are $100 annually or $125 monthly
- Two-seat minimum
Claude API
$10 input / $50 output per 1M tokens
Usage-based pricing for developers, with separate prompt-caching rates.
- Five-minute cache writes: $12.50 per million tokens
- One-hour cache writes: $20 per million tokens
- Cache hits: $1 per million tokens
- US-only inference carries a 1.1x multiplier
Pricing checked . Check current pricing at the source ↗
Assessment
Claude Fable strengths and limitations
Where it stands out
- Anthropic's highest-capability generally available model
- Strong fit for long-horizon agents and difficult repository-scale coding
- One-million-token context with up to 128,000 output tokens
- Broad agent tooling, including memory, code execution, compaction, and programmatic tool use
- Available across Claude apps, the Anthropic API, and major cloud platforms
What to consider
- API use costs twice as much as Claude Opus 4.8 and substantially more than Sonnet
- Responses are slower than Anthropic's lower-cost models
- Adaptive thinking is always on and cannot be disabled
- Stricter safety classifiers can decline harmless requests, so applications need explicit refusal handling
- Fable requires 30-day data retention and does not support zero-data-retention arrangements
Compare
Claude Fable alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Consumer
Claude Opus 4.8
A strong complex-agent and coding model at half Fable's standard API price.
Explore Claude Opus 4.8 →Consumer
Claude Sonnet 5
A faster and much less expensive Claude model for everyday production workloads.
Explore Claude Sonnet 5 →Business Operations
ChatGPT
A broader end-user AI workspace with a large ecosystem of tools and integrations.
Explore ChatGPT →Consumer
Gemini 3.1 Pro
A competing frontier model for advanced reasoning and multimodal work.
Explore Gemini 3.1 Pro →Questions
Claude Fable FAQs
What is Claude Fable 5?
Claude Fable 5 is Anthropic's highest-capability generally available model. It is designed for difficult, long-running coding, research, agent, and knowledge-work tasks.
How much does Claude Fable 5 cost?
The API costs $10 per million input tokens and $50 per million output tokens. Consumer access requires a paid Claude plan, but included usage and credit rules differ by plan.
Is Claude Fable available on the free Claude plan?
No. Anthropic currently limits Fable access to paid Claude plans, usage-based Enterprise accounts, the API, and supported cloud marketplaces.
What is the difference between Claude Fable 5 and Claude Mythos 5?
They share the same underlying capabilities, but Fable is the generally available model with stricter safety safeguards. Mythos is restricted to approved Project Glasswing participants and is not a general consumer option.
How large is Claude Fable's context window?
Fable supports a 1 million-token context window and up to 128,000 output tokens.
How should an API app handle a Fable refusal?
Check the response stop reason even when the HTTP request succeeds. A refusal uses HTTP 200, so the application should route to a safe fallback or ask the user to revise the request.
Does Claude Fable support zero data retention?
No. Anthropic's current documentation says Fable requires 30-day data retention and is not available under zero-data-retention arrangements.
Bottom line
Our Claude Fable verdict
Claude Fable 5 is the model to test when a project is too difficult or long-running for ordinary production models. Its capability, context, and agent tooling are exceptional, but the high token price, slower responses, stricter refusals, and retention requirement make Opus or Sonnet the better default for routine work.
Visit Claude Fable website ↗Similar Consumer Tools
View all Consumer AI tools →
NVIDIA Nemotron 3 Ultra
Nemotron 3 Ultra - Nvidia’s open 550B reasoning model for agents
freddy
Freddy - Plug your wearables, CGMs, power meters, and gym apps straight into any AI agent that speaks MCP
MiniMax M3
MiniMax M3 - Open-weight model with 1M context and computer use
Fusion
Fusion - OpenRouter's new tool that fuses several models into panels to achieve frontier performance
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.