Implement Automatic Model Failovers for Refused Requests
Ensure continuous uptime by setting up automatic model fallbacks for refusals.
Implement the 'fallbacks' API parameter to automatically re-run prompts on a secondary model if the primary model issues a safety refusal.
The Scenario
Your AI application is being used by diverse users and occasionally hits safety 'refusal' stop reasons. You want to ensure the user experience isn't interrupted by a hard error.
Before & after
When a request was blocked by a safety classifier, the application would simply fail or require complex 'try-catch' logic to manually retry with a different model. Designing this error handling manually usually took 15–20 minutes.
By using the 'fallbacks' parameter in the Claude API, the system automatically redirects a blocked request to your chosen backup model. This fail-safe mechanism takes 1 minute to add to your API call code.
The Prompt
Review my existing Claude API request code and show me how to implement the 'fallbacks' parameter to ensure that if [MODEL_NAME] refuses a request, it automatically fails over to [BACKUP_MODEL]. [PASTE_CODE_HERE]
Claude Fable 5 introduces an opt-in 'fallbacks' parameter. If the primary model refuses a request due to safety filters, the system can automatically re-run the prompt on a different model (like a smaller, faster one) without requiring a new round-trip request from your application.
Source
Claude Platform - Claude API Docs"An opt-in fallbacks parameter... re-runs refused requests on another model, billed at the fallback model's rates."