Anthropic Launches Opus 5, Cheaper and Less Restrictive AI Model

AI company Anthropic has launched Opus 5, the newest iteration of its large model. While smaller than the previously released Fable 5, Opus 5 is positioned to be both more cost-effective and less restrictive, suggesting it will be favored for a wider array of applications.

Benchmark testing indicates that Opus 5 outperforms Fable 5 in several key areas. This release follows closely on the heels of Opus 4.8, which became available in late May, with other models like Mythos 5, Fable 5, and Sonnet 5 having been launched in June.

Anthropic highlighted Opus 5's enhanced capabilities in verifying its work and iterating precisely until successful. An example provided involved the model independently developing a computer vision pipeline based on an incomplete prompt, showcasing its advanced reasoning.

A significant advantage of Opus 5 is its reduced data retention policy compared to Fable. Unlike Fable and Mythos, which adhere to a 30-day data retention policy that has raised privacy concerns, Opus 5 offers a less stringent approach. However, it still incorporates safeguards for sensitive cybersecurity tasks, such as prohibiting exploit generation while allowing source code vulnerability analysis for defensive purposes.