Opus 5 Ships: Better Benchmarks, 85% Fewer Classifier Hits Than Fable 5
Anthropic released Opus 5 today. It outperforms Fable 5 on multiple benchmarks, costs less, and triggers safety classifiers far less often. The flagship model also avoids the 30-day data retention policy that applies to Fable and Fable 5.
A Crowded Summer
The Claude 5 family had a busy June. Mythos 5, Fable 5, and Sonnet 5 all shipped that month. Opus 5 arrives two months after Opus 4.8, which became available May 28. Haiku has not received a 5-series upgrade as of today's launch.
Safety Classifiers Fire 85% Less Than Fable 5
Anthropic expects Opus 5 safety classifiers to engage 85% less often than they do on Fable 5. For API users who have been hitting refusals on Fable 5, that number is the headline.
The remaining restrictions follow a documented logic. Opus 5 will not assist with scanning vulnerabilities in compiled binaries. It will assist with vulnerability research in source code. Whether that line holds up in practice is an empirical question, but it is at least specific.
No 30-Day Data Retention
Fable and Fable 5 carry a 30-day data retention policy. Opus 5 does not. For deployments with data sensitivity requirements, this is a meaningful difference from the Fable line.
Automatic Fallbacks
Anthropic also shipped Automatic Fallbacks as a beta feature. When a prompt triggers a safety classifier, the request routes to a less capable model instead of returning an error. API users opt in.
The result is a functional response rather than a hard refusal. Whether the fallback model's output is actually useful depends entirely on what triggered the classifier in the first place.
The Positioning Is Odd
Anthropic's most capable model is now also its cheaper, less restricted one. Fable 5 was built with safety-forward defaults. Opus 5 is built differently. Anthropic has not offered a public explanation for why the flagship and the safety-focused model diverge this sharply on classifier behavior.
One possibility is that Fable's posture serves specific institutional deployment contexts. Opus 5 may be targeting developers and researchers for whom friction is the primary obstacle. That is speculation, not Anthropic's stated rationale.
Source: Techcrunch