source: techcrunch ai: anthropic launches opus 5
level: technical
anthropic launched its opus 5 model on friday, the latest version of its heavyweight series. the model is smaller than fable 5 but costs less and has fewer restrictions, making it more practical for most uses. opus 5 outperforms fable 5 on multiple benchmarks, including tasks where it verifies its own work and iterates carefully. it was released just two months after opus 4.8, while other models in the 5 series like mythos, fable, and sonnet launched in june. only the lightweight haiku model still awaits an upgrade.
opus 5 avoids many of the restrictions that have affected fable 5. it is not subject to the 30-day data retention policy that applies to fable and mythos, which had worried some privacy-conscious users. safeguards remain for cybersecurity tasks, such as blocking exploit generation and penetration testing on software binaries, though source code vulnerability searches are allowed for defensive purposes. anthropic expects safety classifiers to trigger 85% less often for opus 5 compared to fable 5, reflecting a lighter touch for this less capable model.
anthropic also introduced a beta feature called automatic fallbacks. when a prompt triggers the safety classifier, this tool routes the request to a less powerful model instead of returning an error. api users who opt in will receive a functional response, reducing disruption from safeguards.
why it matters: opus 5 offers a more practical and less restricted option for ai developers, balancing performance with lower cost and fewer privacy concerns.