TheModelDesk

Anthropic launched Opus 5 on Friday, July 24, 2026, the latest iteration of its heavyweight model line. Although smaller than Fable 5, it is cheaper and subject to fewer usage restrictions, which Anthropic expects will make it the better choice for most use cases. According to the announcement benchmarks, Opus 5 actually outperforms Fable 5 on several tests.

The release comes just two months after Opus 4.8 (May 28). Mythos 5, Fable 5, and Sonnet 5 all shipped in June, leaving the lightweight Haiku as the only model still awaiting a 5-series upgrade.

Anthropic highlights that Opus 5 is much better at verifying its own work and iterating until it succeeds. In one cited benchmark, the model built its own computer vision pipeline in response to an incomplete prompt.

Opus 5 is not covered by the 30-day data retention policy that applies to Fable and Mythos — a policy that had drawn criticism from privacy-conscious users. Safeguards remain for cybersecurity tasks such as exploit generation and penetration testing: the model may not scan binaries for vulnerabilities, but it may search source code, since the latter is more often defensive. Anthropic expects its safety classifiers to trigger 85% less often on Opus 5 than on Fable 5.

A new beta feature called Automatic Fallbacks aims to soften the impact of those classifiers: when a prompt trips a safety filter, the request is automatically routed to a less capable model, so API users receive a functional answer instead of an error.