Anthropic announced Opus 5 on Friday, the latest entry in its heavyweight model line. Opus 5 is smaller than Fable 5 but is positioned as both less restrictive and less expensive, which Anthropic says will make it preferable for many use cases.
Performance and benchmarks
According to Anthropic, Opus 5 outperforms Fable 5 on several benchmarks cited in the announcement. The company emphasized that Opus 5 is "much stronger at verifying its work and iterating carefully until it succeeds," and noted an example in which Opus 5 wrote its own computer vision pipeline in response to an incomplete prompt during benchmark testing.
Release cadence and model family
Opus 5 arrives only 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 yet to receive a 5-series upgrade.
Data handling and restrictions
Crucially, Opus 5 is not subject to the 30-day data retention policy that covers Fable and Mythos; that policy had prompted concerns among some privacy-conscious users. At the same time, Anthropic maintains meaningful safeguards, especially for cybersecurity-related tasks such as exploit generation and penetration testing. For example, Opus 5’s protections prevent it from being used to scan a software binary for vulnerabilities, while searching source code for vulnerabilities is permitted because it is more likely to be used defensively.
Classifiers and Automatic Fallbacks
Anthropic expects safety classifiers to trigger about 85% less often for Opus 5 than for Fable 5, reflecting a lighter touch for the less-capable model. The company is also introducing a new tool to reduce disruption when safeguards do engage: a beta feature called Automatic Fallbacks. When enabled, Automatic Fallbacks will automatically route requests that trigger the safety classifier to a less powerful model, so API users receive a functional response instead of an error message.
Why it matters
The Opus 5 release shows Anthropic balancing performance, cost, and safety: a smaller, cheaper, and in some ways less restricted model that claims improved self-checking abilities, while retaining protections for sensitive tasks. The Automatic Fallbacks feature aims to smooth the user experience when safety classifiers intervene.
