Anthropic launches Opus 5: cheaper and less restrictive than Fable

Anthropic launched Opus 5, the newest version of its heavyweight model, on Friday. It is smaller than Fable 5 but cheaper and less restrictive, making it preferable for most use cases. Opus 5 outperforms Fable 5 on several benchmarks included in the announcement. The model arrived only two months after Opus 4.8, while Mythos 5, Fable 5, and Sonnet 5 launched in June, leaving only Haiku waiting for a 5-series upgrade. Anthropic emphasized that Opus 5 is stronger at verifying its work and iterating carefully, citing benchmark testing where it wrote its own computer vision pipeline from an incomplete prompt. Opus 5 is free from many restrictions that affected Fable, including the 30-day data retention policy. Safeguards remain for cybersecurity tasks like exploit generation and penetration testing; for instance, Opus 5 prevents scanning for vulnerabilities in a software binary but permits source code vulnerability searches. Anthropic expects these classifiers to engage 85% less often for Opus 5 than for Fable 5. A new beta feature called Automatic Fallbacks allows users to route requests to a less powerful model when safety classifiers trigger, providing a functional response instead of an error.