Claude Fable 5.1 is Anthropic's new top generally available model. It is designed for the assignments that expose weaker agents: hours of tool use, large codebases, multistep research, document-heavy analysis, failures that require recovery, and deliverables that must survive human review.
Anthropic released Fable 5.1 on September 1, 2026. It is available to Claude Pro, Max, Team, and Enterprise users and through the Claude API and major cloud platforms. The API model ID is claude-fable-5-1.
The useful positioning
Fable 5.1 is an escalation model for demanding, long-horizon work.
Anthropic's own model-selection guide recommends starting most workloads with Opus 5, then moving to Fable 5.1 when demanding reasoning or agent work still falls short in your evaluations.
What changed from Fable 5?
Fable 5.1 keeps the same standard input and output token prices while strengthening long-running agentic coding, multistep research, and work across documents, spreadsheets, and slides. Cache reads fall to one quarter of the previous price: $0.25 per million tokens. Anthropic estimates that this can reduce typical workload cost by about 25% and highly agentic workload cost by up to roughly 45%, depending on cache use.
Longer horizon
Plan, act, recover, and continue across workflows that take hours rather than one answer.
Root causes
Anthropic emphasizes fixing the underlying problem instead of accepting an easy-looking shortcut.
Visible progress
Readable updates between tool calls make long sessions easier for a person to supervise.
Cheaper cache reads
Repeated project context can cost less when an application is designed to reuse cached prefixes.
Where Fable 5.1 is strongest
Repository-scale coding
Fable 5.1 is meant for features that span a codebase, difficult code review, performance work, visual implementation, and autonomous sessions that may continue for days. It can generate tests, use vision to compare output with a design goal, and keep working through failed steps.
Research that becomes a deliverable
The model is positioned for multistage research and analysis that ends in a reviewable document, spreadsheet, or presentation. That distinction matters: collecting sources is only one phase; selecting evidence, resolving conflicts, structuring the argument, and formatting the final artifact require sustained context.
Document-heavy professional work
Fable 5.1 accepts text and images and can interpret charts, diagrams, tables, and pages inside files and PDFs. This can help with finance, analytics, legal, architecture, and other workflows where the evidence is visual as well as textual.
Claude Fable 5.1 specifications
| API model ID | claude-fable-5-1 |
|---|---|
| Context window | 1,000,000 tokens |
| Maximum output | 128,000 tokens |
| Reliable knowledge cutoff | June 2026 |
| Thinking | Adaptive, always on; default effort high |
| Standard API price | $10 input / $50 output per million tokens |
| Cache read | $0.25 per million tokens |
How to prompt a long-running Fable task
A powerful model still needs a controlled assignment. Give it the outcome, working environment, authority, checkpoints, and evidence. For large code changes, specify whether you want focused edits or broader rewrites; Anthropic notes that Fable 5.1 can be more willing than Fable 5 to rewrite a full file.
Work on this as a long-horizon project. Goal: [finished, reviewable outcome] Inputs: [repositories, files, sources, and tools] Constraints: [scope, compatibility, style, budget, and deadlines] Protected behavior: [what must remain unchanged] Before acting: 1. inspect the environment; 2. identify missing evidence and dependencies; 3. create a phased plan with verification after each phase. While working: - report concise progress between major tool calls; - recover from failures before changing direction; - prefer targeted edits unless a rewrite is justified; - finish with tests, evidence, remaining risks, and a handoff.
The safeguard and retention details you should not skip
Anthropic says Fable 5.1 uses additional safeguards for cybersecurity, biology, and chemistry. Some flagged requests can be routed to less capable Claude models, and API applications need to account for the fallback behavior. Using Fable also requires 30-day data retention for safety monitoring by default, with different arrangements for eligible enterprise customers.
Those details affect product design. Teams handling sensitive code or regulated information should confirm the applicable retention terms, cloud configuration, and fallback behavior before moving production traffic.
Who should actually use Fable 5.1?
✓Teams whose hardest real evaluations still fail on Opus 5 at higher effort.
✓Agents that must stay coherent across hours of tools, files, and dependent decisions.
✓Developers tackling repository-wide features, rare failures, or demanding performance work.
✓Researchers turning mixed evidence into a polished, reviewable professional artifact.
Our practical verdict
Fable 5.1 is compelling when persistence and judgment across a long workflow create measurable value. Start from representative evaluations, include review time and cache economics, and pay for Fable where the final result proves that Opus or Sonnet is not enough.