Claude Opus 5 is now available, offering near frontier intelligence at half the price of Claude Fable 5. It sets a new state-of-the-art on coding and knowledge work evaluations like Frontier-Bench and GDPval-AA, while remaining behind Mythos 5 on cybersecurity.

Opus 5 is designed for daily use, operating more efficiently than previous models. It provides greatly improved performance for the same cost as Opus 4.8, with customizable effort settings to optimize for intelligence or conserve tokens. The model excels in software engineering, surpassing all others on Frontier-Bench v0.1 and nearing Fable 5's performance on Cursor Bench 3.2 at a lower cost.

In knowledge work and problem-solving, Opus 5 achieves significantly higher scores on evaluations like ARC-AGI 3 and Zapier Automation Bench. It also demonstrates enhanced capabilities in scientific research, particularly in organic chemistry and protein-related tasks, and produces stronger visual outputs. Opus 5 exhibits greater agency and thoroughness, successfully completing complex tasks that other models could not, such as reconstructing a machine part from a drawing without direct viewing access and fixing subtle bugs in software.