Claude Fable 5 in OffloopStay with an ambitious project.
Use Claude Fable 5 in Offloop for extended engineering, analysis and visual reasoning.
A brief to try · Example
Implement a multi-stage project
Break this project into milestones. Implement one milestone at a time, inspect the output and preserve a concise progress record.
Choose in your agent settings: Claude Fable
Where Claude Fable 5 fits
Our pick for demanding projects that benefit from repeated inspection and revision.
01
Implement a multi-stage project
Long-horizon engineering is a central focus of the release and its evaluations.
02
Reconcile charts and documents
Anthropic reports improvements in visual and analytical work.
03
Refine an interactive prototype
Use visible output and explicit feedback to guide successive revisions.
Our editorial assessment · This page covers Fable 5, not 5.1. Anthropic describes safeguards that may route some requests to Opus 4.8; published results depend on those conditions.
Read the capability source ↗Watch a real evaluation
A coding review of the original Fable 5. The creator promotes Depot in the description; treat the demonstrations as qualitative evidence.
ForrestKnight · Independent creator · External setup, not Offloop
Published evaluations
Selected results for Claude Fable 5. Source checked November 5, 2026; these are not Offloop measurements.
SWE-bench Pro · max
80.4%
Fable comparison value reproduced in the Grok 4.5 launch table.
Terminal-Bench 2.1 · max
84.3%
Provider-reported comparison; not an Offloop measurement.
SWE Marathon · max
24.0%
Pass@1 in the same comparison; below Opus 4.8 on this particular test.
Different task sets, model snapshots and harnesses are not directly comparable. Scores are evidence for a trial, not a guarantee of your result.
Before you start
Give your agent a real task.
Choose Claude Fable, connect the tools you need, and define what a useful result looks like.