Claude Sonnet 5 System Card (detailed metrics) [pdf] (www-cdn.anthropic.com)

Anthropic has announced Claude Sonnet 5, the latest iteration in its Sonnet model lineup, which builds upon the enhancements of Claude Sonnet 4.6. This upgrade showcases improvements in areas such as coding, agentic performance, and professional task execution. Notably, while Sonnet 5 exhibits significant upgrades in its handling of alignment measures and agentic safety—particularly in robustness against prompt injections—it does not surpass the capabilities of Anthropic's higher-class models, Opus and Mythos. The model maintains a low alignment risk, although slightly higher than its predecessor, and demonstrates limited advancement in automated AI research capabilities. The introduction of Claude Sonnet 5 is significant for the AI/ML community as it bridges the gap between the affordability of Sonnet models and the advanced capabilities typical of higher-tier models. With enhancements in the timing of responses to potentially harmful requests and overall performance metrics, this model shows a commitment to responsible AI deployment. The system card details rigorous evaluations of its safeguards and welfare, emphasizing a dual focus on performance and ethical behavior. This positions Sonnet 5 as a robust option for developers seeking a balance between capability and responsible AI use.