Claude Fable 5

Anthropic's widely released flagship for demanding reasoning and long-horizon agentic work.

Context1M

ModalityText · Image

AccessAPI · Cloud

ReasoningAdaptive

Where it fits.

A fit for complex, open-ended work that must preserve task state over long runs rather than optimize for raw latency. Anthropic describes Fable 5 as its most capable widely released model for long-running agents. It has a 1M default context, 128K maximum output and always-on adaptive thinking across major cloud channels.

Core capabilities

1M context and up to 128K output
Always-on adaptive thinking for long-horizon work
Text and image understanding with agentic tool use
Availability through Claude API and major managed clouds

Best-fit workflows

Long-horizon autonomous work
Complex reasoning and planning
High-value enterprise workflows

04 / OPERATIONS

From evaluation
to production.

Deployment notes

Use complete, unmodified thinking blocks with tool results when preserving agent state. Build checkpoints and resumability around long runs, and compare Fable against Opus or Sonnet on a task-specific eval before paying the premium.

Watch before choosing

Long context is not lossless memory; retrieval, citations and staged summaries still matter. Fable is the premium, slower tier. Non-default sampling parameters are restricted, and retained thinking blocks consume context even when they improve multi-step continuity and caching.

Family & versions

Claude Fable 5Widely released flagship
Claude Opus 5Complex agents
Claude Sonnet 5Frontier at scale
Claude Haiku 4.5Fast

Specifications and positioning checked against provider documentation.