Anthropic launches Opus 5 | TechCrunch

On Friday, Anthropic launched its Opus 5 model, the latest version of its long-standing heavyweight model. Although smaller than Fable 5, the model will be both cheaper and less restrictive than Fable, making it likely preferable in most use cases.

Notably, Opus 5 outperforms Fable 5 on a number of benchmarks included in the announcement.

Opus 5 launches just two months after Opus 4.8, which became available on May 28. Mythos 5, Fable 5 and Sonnet 5 all launched in June, leaving only the lightweight Haiku model still awaiting an upgrade to the 5 series.

In a message announcing the new modelAnthropic highlighted that Opus 5 was “much stronger at verifying its work and carefully iterating until it succeeds,” citing benchmark tests, in which Opus 5 wrote its own computer vision pipeline in response to an incomplete prompt, among other examples.

Crucially, Opus 5 is also free from many of the limitations that have plagued Fable since its release. Like its predecessor, Opus 5 is not subject to the 30 day data retention policy that covers Fable and Mythos, raising concerns among some privacy-conscious users.

There are still meaningful safeguards for Opus, especially around cybersecurity tasks such as exploit generation and penetration testing. For example, Opus 5 safeguards prevent it from being used to scan for vulnerabilities in a software binary, although it is allowed to look for vulnerabilities in the source code, as the latter task is more likely to be used for defensive purposes.

Broadly speaking, Anthropic expects these classifiers to be involved 85% less often in Opus 5 than in Fable 5, reflecting the lighter touch given to the less capable model.

Anthropic is also introducing a new tool to make security measures less disruptive if they do occur. Users can now sign up for a beta feature called Automatic Fallbacks, which automatically redirects requests to a less powerful model when a prompt triggers the safety rating. The result is that API users with the setting enabled will receive a functional response instead of an error message.