Claude Opus 4.7 Model Card (anthropic.com)

Anthropic has announced the release of Claude Opus 4.7, an advanced large language model that offers notable improvements over its predecessor, Claude Opus 4.6, positioning it as the most capable general-access model available from the company. While it does not surpass the more powerful Claude Mythos Preview, which has limited user access, Opus 4.7 enhances key functionalities across evaluations, particularly in software engineering and real-world professional tasks. The new model also addresses safety and security concerns with improved safeguards and a robust alignment profile, demonstrating low rates of hallucinations and effective adherence to its guidelines. The release is significant for the AI/ML community as it maintains a focus on responsible scaling, indicating that catastrophic risks remain minimal. Evaluations reveal that while Opus 4.7 has similar cyber capabilities to its predecessor, it introduces enhanced protections against malicious agentic requests, demonstrating greater robustness in prompt injection scenarios. Furthermore, the model exhibits consistent performance on essential metrics like user safety and election integrity. These advancements contribute to the ongoing discourse around the balance of AI capabilities and safety measures, underscoring the importance of responsible AI development in a rapidly evolving landscape.