On Friday, Anthropic released the Opus 5, the newest version of its long-running heavyweight model. Although smaller than the Fable 5, the model will be both cheaper and less restrictive than the Fable, making it likely preferable in most use cases.
It’s worth noting that the Opus 5 outperforms the Fable 5 in some benchmarks included in the announcement.
Opus 5 comes just two months after Opus 4.8, which became available on May 28. The Mythos 5, Fable 5 and Sonnet 5 all launched in June, leaving only the lightweight Haiku model still waiting for an upgrade to the 5 series.
In a post announcing the new modelAnthropic emphasized that Opus 5 was “much stronger at verifying its work and carefully iterating until it succeeds,” citing the benchmark test, in which Opus 5 wrote its own line of computer vision in response to an incomplete prompt, among other examples.
Most importantly, Opus 5 is also free of many of the limitations that have plagued Fable since its release. Like its predecessor, Opus 5 is not subject the 30-day data retention policy covering Fable and Mythos, which had raised concerns among some privacy-conscious users.
Important safeguards remain for Opus, particularly around cybersecurity tasks such as exploit generation and penetration testing. For example, Opus 5’s safeguards prevent it from being used to scan for vulnerabilities in a software binary, although searching for vulnerabilities in the source code is allowed, as the latter work is more likely to be used for defensive purposes.
In general, Anthropic expects these classifiers to engage 85% less often for Opus 5 than for Fable 5, a reflection of the lighter touch given to the less capable model.
Anthropic is also launching a new tool to make safeguards less annoying when they get involved. Users can now select a beta feature called Automatic Fallbacks, which will automatically route requests to a less robust model when a prompt triggers the security classifier. The result is that API users with the setting enabled will receive a functional response instead of an error message.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
