Introducing Claude Opus 5

Claude Opus 5 is a new state-of-the-art AI model designed for everyday use in coding, knowledge work, and scientific research. It matches the pricing of its predecessor, Opus 4.8, while delivering greatly improved performance across software engineering, problem-solving, and financial or legal workflows. Users can adjust effort settings to optimize for intelligence or speed, and the model is also available in a faster beta mode at twice the base price.

The model excels at agentic tasks, showing strong capabilities in verifying its own work, debugging complex codebases, and handling multi-step reasoning. In evaluations, Opus 5 leads on benchmarks like Frontier-Bench and CursorBench, approaches the performance of higher-tier models at a fraction of the cost, and generates significantly improved visual artifacts and scientific analyses, particularly in organic chemistry and protein biology.

In safety and alignment testing, Opus 5 is the most aligned model to date, adhering well to safety standards and exhibiting low rates of deceptive behavior. While it can identify software vulnerabilities nearly as well as the specialized Mythos 5 model, it falls significantly behind in developing exploits. The model features tailored safeguards, including relaxed cyber classifiers for finding source code vulnerabilities and specialized access programs, alongside automatic API routing fallbacks for flagged requests.