You Didn’t Get the AI Model You Paid For

Editor's take

Users attempting to access Anthropic's "claude-fable-5" model via API are instead receiving responses from "claude-opus-4-8," with no explicit error or indication of the substitution.

This discrepancy is significant because it directly impacts user expectations and potential billing, especially for those relying on specific model capabilities and performance characteristics. It raises questions about transparency in API offerings and the potential for unseen cost or quality shifts, affecting developers and businesses integrating these LLMs into their applications.

Future observations should focus on whether Anthropic clarifies its model versioning and API response fidelity, and if other major LLM providers exhibit similar practices. Understanding the rationale behind these model substitutions, whether for load balancing, A/B testing, or resource management, will be crucial for assessing the reliability of commercial LLM deployments.