Anthropic released Opus 5 on July 25, 2026, the latest version of its heavyweight AI model, positioning it as a cheaper and less restrictive alternative to Fable 5 that the company expects will be preferable in most use cases.
The new model arrives just two months after Opus 4.8, which launched on May 28. Mythos 5, Fable 5, and Sonnet 5 had all launched in June, leaving Opus 5 as the latest addition to Anthropic’s 5-series lineup. Only the lightweight Haiku model has yet to receive a 5-series upgrade.
Despite being smaller than Fable 5, Opus 5 outperforms it on a number of benchmarks cited in Anthropic’s announcement. The company highlighted the model’s ability to verify its own work and iterate until it succeeds, pointing to a benchmark in which Opus 5 wrote its own computer vision pipeline in response to an incomplete prompt.
A key distinction for Opus 5 is its reduced data retention restrictions. Like its predecessor, Opus 5 is not subject to the 30-day data retention policy that applies to Fable and Mythos — a policy that had raised concerns among privacy-conscious users. Anthropic expects its safety classifiers to engage 85% less often for Opus 5 than for Fable 5, reflecting the lighter restrictions placed on the less capable model.
Some safeguards remain in place, particularly around cybersecurity tasks. Opus 5 cannot be used to scan for vulnerabilities in software binaries, though it is permitted to search for vulnerabilities in source code, which Anthropic describes as more likely to serve defensive purposes.
Alongside the launch, Anthropic introduced a beta feature called Automatic Fallbacks. When a prompt triggers a safety classifier, the feature automatically routes the request to a less powerful model rather than returning an error message, giving API users a functional response instead of a refusal.
Source: TechCrunch