Introducing Claude Opus 5

Anthropic has released Claude Opus 5, a new artificial intelligence model that currently leads the Artificial Analysis leaderboard, outperforming Claude Fable 5 while costing half the price. It maintains the same base pricing as the previous Opus 4.8 version and continues to offer a fast mode at twice the standard cost.

In evaluation testing, the model demonstrated advanced proactive problem-solving capabilities. When assigned a Frontier-Bench task to write code for a 3D FreeCAD model of a machine part without direct access to the drawing, Opus 5 independently wrote its own computer vision pipeline to extract the geometry from raw pixels and successfully reconstructed the part.

The model also shows significant improvements in cybersecurity capabilities, approaching the performance of Mythos 5 in identifying vulnerabilities. However, Anthropic deliberately avoided training Opus 5 on cyber tasks, and the model remains substantially behind Mythos 5 in exploiting those vulnerabilities to create material cyber threats.

Alongside the model release, Anthropic published a prompting guide, and Thariq Shihipar released a guide on context engineering for Claude 5 generation models.