Anthropic Launches Claude Opus 5 With Premium AI Performance
Anthropic has released Claude Opus 5, a highly efficient artificial intelligence model that delivers near-frontier capabilities at half the cost of the flagship Claude Fable 5.
The new system maintains the pricing structure of Opus 4.8, operating at $5 per mln input tokens and $25 per mln output tokens, while providing significant performance upgrades. On the ARC-AGI 3 evaluation, which tests novel problem-solving capabilities, Opus 5 tripled the score of the next-best model. Furthermore, on the OSWorld 2.0 computer use benchmark, it outperformed every competitor at any given cost.
Claude Opus 5 excels heavily in software engineering, knowledge work and complex scientific research. On the Frontier-Bench evaluation, the model surpassed all competing systems and more than doubled the performance of Opus 4.8. Early-access users noted its unique ability to write complex code, build custom test harnesses and independently verify its own work. In one instance, the model successfully wrote a computer vision pipeline from scratch to reconstruct a 3D machine part from raw pixels after competing models failed repeatedly.
Breakthroughs in life sciences
In the scientific sector, the model demonstrated substantial improvements across organic chemistry, bioinformatics and structural biology. It vastly outperformed previous iterations in predicting molecular structures from spectroscopy data and analysing how protein sequence variations affect overall functionality. The system also generates vastly superior visual outputs, allowing users to create detailed interactive cell illustrations and aerodynamic flow visualisations.
Business leaders from companies like Zapier and Box reported that Opus 5 handles long-horizon tasks with unprecedented reliability. The model can process raw account-health workbooks, flag at-risk clients and alert the appropriate personnel without losing track of the core objective.
Rigorous cybersecurity safeguards
Internal audits indicate Claude Opus 5 is the company’s most aligned and secure model to date. It adheres strictly to Claude’s Constitution, exhibits minimal deceptive behaviour and strongly resists external misuse attempts.
While the system shows advanced capability in identifying software vulnerabilities, Anthropic intentionally restricted its capacity to exploit those weaknesses. Built-in safeguards prevent binary-based vulnerability scanning, penetration testing and exploit generation to mitigate global cybersecurity threats. Any API requests triggering these safety classifiers will automatically revert to the older Opus 4.8 model unless users possess specific enterprise security clearances through the Cyber Verification Program.
The release also introduces beta updates for developers, allowing mid-conversation tool changes without invalidating prompt caches. Opus 5 is available immediately across all Anthropic platforms, including a Fast mode option that provides processing speeds 2.5 times quicker than the default rate.
