On Friday, Anthropic launched its newest heavyweight model, Opus 5. This version not only outperforms Fable 5 in several benchmarks but also offers a friendlier price tag and fewer restrictions for users.

Unlike its predecessor, Opus 5 is free from the 30-day data retention policy that has caused privacy concerns among some users of Fable and Mythos. However, it still comes with certain safeguards, particularly when it comes to cybersecurity tasks such as exploit generation and penetration testing.

Broadly speaking, these classifiers are expected to engage 85% less often for Opus 5 than they do for Fable 5. Anthropic has also introduced a new tool called Automatic Fallbacks, which automatically routes requests to a less powerful model when a prompt triggers the safety classifier. This ensures that API users with this feature engaged will receive a functional response rather than an error message.

With Opus 5's improved performance and reduced restrictions, it is likely to be preferred in most use cases over Fable 5. However, users should still exercise caution when handling sensitive data as the model is not entirely without its limitations.