Claude Opus 5: Anthropic's New Model Balances Performance and Safety

Anthropic has released Claude Opus 5, a model designed to deliver practical performance comparable to Claude Fable 5 while offering twice the speed at half the cost. Compared to its predecessor Claude Opus 4.8, the new model shows substantial improvements across coding, computer use, and complex knowledge work. Benchmark results from third-party evaluators position it among the strongest models currently available. In high-risk domains such as cybersecurity, Claude Opus 5 remains significantly weaker than Claude Mythos 5, reflecting deliberate training choices to limit dangerous capabilities. Nevertheless, it surpasses Opus 4.8 substantially across these same tasks. Anthropic has deployed improved safeguards that trigger 85% less frequently than earlier versions, reducing false positives while maintaining security effectiveness. A significant change permits source code vulnerability analysis at all access levels while continuing to block compiled binary analysis, reflecting the defensive value of code review versus the offensive nature of binary exploitation. The model demonstrates dramatic improvements in prompt injection resistance, reducing attack success rates from 5.5% to 2.0% in computer use scenarios—outperforming other major models by an order of magnitude. Overall alignment and agentic safety metrics show consistent improvements, establishing Claude Opus 5 as a practical option for deployments requiring both capability and responsible behavior.