Anthropic Debuts Opus 5 to Address Fable's Constraint Issues

Opus 5 demonstrates superior performance across several key benchmarks, proving itself more adept at self-verification and iterative problem-solving. In internal testing, the model successfully constructed a complex computer vision pipeline from an incomplete prompt, showcasing its improved autonomy. Unlike Fable and Mythos, Opus 5 remains exempt from the controversial 30-day data retention policy, a shift likely to appease privacy-focused enterprise clients.

While the model retains essential cybersecurity safeguards—specifically blocking binary-level vulnerability scanning—it allows for source code analysis to encourage defensive development. Anthropic anticipates that its safety classifiers will trigger 85% less frequently than they do for Fable 5. To further minimize disruption, the company introduced Automatic Fallbacks, a beta feature that reroutes blocked requests to smaller, less restricted models, ensuring API users receive functional outputs rather than error messages.