Anthropic launches Opus 5, positioning it as a cost-effective alternative to Fable 5
Released two months after Opus 4.8, the latest iteration from Anthropic offers improved benchmark performance and reduced restrictions compared to its Fable 5 counterpart.
Anthropic has officially launched Opus 5, the newest iteration of its heavyweight artificial intelligence model series. The release arrives two months after Opus 4.8, which became available on 28 May, and follows the June debut of other models in the 5 series, including Mythos 5, Fable 5, and Sonnet 5. The lightweight Haiku model remains the only entry in the current lineup yet to receive an upgrade to version 5.
According to the announcement, Opus 5 is smaller than Fable 5 but is designed to be cheaper and less restrictive, a combination Anthropic suggests makes it preferable for most use cases. Despite its smaller size, the model outperforms Fable 5 on several benchmarks included in the launch details. Anthropic highlighted improved capabilities in work verification and iteration, citing examples where the model successfully wrote its own computer vision pipeline in response to incomplete prompts.
A significant differentiator for Opus 5 is its data handling policy. Unlike Fable 5 and Mythos 5, which are subject to a 30-day data retention period that has raised concerns among privacy-conscious users, Opus 5 is not subject to this policy. This change addresses previous friction points regarding data storage for users of the company’s most capable models.
While the model is less restrictive overall, specific safeguards remain in place for cybersecurity tasks. Opus 5 cannot be used to scan software binaries for vulnerabilities, although it is permitted to search for vulnerabilities in source code, a distinction Anthropic notes is more likely to be used for defensive purposes. The company expects safety classifiers to engage 85% less frequently for Opus 5 than for Fable 5, reflecting a lighter touch on restrictions for the less capable model.
To mitigate disruptions when safety triggers do activate, Anthropic is introducing a beta feature called Automatic Fallbacks. This tool allows API users to automatically route requests to less powerful models when a prompt triggers a safety classifier. The feature ensures that users receive functional responses rather than error messages, aiming to maintain workflow continuity while adhering to safety protocols.
