Key capabilities
- Most capable Claude — Anthropic’s top-tier model for reasoning and agentic coding
- Adaptive thinking always on — Smart reasoning built into every request, no configuration needed
- 1M context window — 128K max output tokens
- OpenAI-compatible — Works as a drop-in replacement with the OpenAI SDK
- Safety classifiers — May decline certain requests with
finish_reason: "refusal", not billed when refusals occur before output - Fallback support — Pass
fallbacksto automatically retry refused requests on another model - Streaming — Supports real-time token streaming via SSE
Quick example
Parameters
Claude Fable 5 does not support custom sampling values. Omit
temperature, top_p, and top_k for normal requests. For backward compatibility, temperature: 1 and top_p values from 0.99 to 1 are accepted; any top_k value or other sampling value returns a 400 error. Use prompting to guide model behavior. See Anthropic’s API parameter deprecations.Claude Fable 5 includes safety classifiers that can return
finish_reason: "refusal". You are not billed for refused requests. Use the fallbacks parameter to automatically retry refused requests on other Claude models.API Reference
View the interactive API playground for Claude Fable 5.