Anthropic unveils Claude Opus 5, positioning it as a proactive model with near-frontier performance at reduced cost

Anthropic has released Claude Opus 5, positioning it as a "thoughtful and proactive model" that approaches the frontier intelligence of Claude Fable 5 while costing half the price. The model is currently leading the Artificial Analysis leaderboard, surpassing even Fable 5. Pricing for Opus 5 remains unchanged from Opus 4.8, and it continues to offer a "fast mode" priced at twice the cost of the base model.

Anecdotal evidence from Anthropic’s release post highlights Opus 5’s proactive capabilities. In a Frontier-Bench task, the model was given a drawing of a machine part and tasked with writing code to rebuild it as a 3D FreeCAD model. Despite being intentionally denied direct access to the drawing, Opus 5 responded by writing its own computer vision pipeline to extract geometry from raw pixels and reconstruct the full machine part.

In cybersecurity-related performance, Anthropic states that Opus 5 has been deliberately not trained on cyber tasks. Nevertheless, the model shows substantial improvement in finding cybersecurity vulnerabilities due to general capability gains, coming close to Mythos 5 in vulnerability detection. However, it remains substantially behind Mythos 5 in the ability to exploit those vulnerabilities—i.e., turning detected vulnerabilities into material cyber threats.

Anthropic has published a dedicated prompting guide for Claude Opus 5, and third-party context engineering guidance has also emerged to optimize its use. Early user reports suggest iterative refinement may be necessary to achieve desired outputs, as evidenced by Simon Willison’s attempts to generate a pelican illustration with mixed initial results.