Anthropic Launches Claude Opus 4.7 with Enhanced Software Engineering and Vision Capabilities
The new model solves four coding tasks its predecessor couldn't, with enhanced safeguards for cybersecurity.
Anthropic has officially released Claude Opus 4.7, positioning it as a significant leap forward for complex software engineering and professional creative tasks. The model shows a 13% improvement over Opus 4.6 on Anthropic's internal 93-task coding benchmark, uniquely solving four tasks that stumped both its predecessor and the Sonnet 4.6 model. Early testers highlight its ability to catch logical faults during planning and accelerate execution, making it particularly effective for long-running, multi-step workflows like automations and CI/CD. It also boasts substantially better vision capabilities, producing higher-quality outputs for professional documents, slides, and user interfaces.
A key strategic aspect of the Opus 4.7 release is its role in testing new cybersecurity safeguards. Anthropic has intentionally limited its cyber capabilities compared to the more powerful, restricted Claude Mythos Preview model. Opus 4.7 launches with automated systems to detect and block prohibited or high-risk cybersecurity requests. The learnings from this deployment are intended to pave the way for a future broad release of Mythos-class models. For legitimate security professionals, Anthropic has established a Cyber Verification Program. The model is available today across all Claude products, its API, and major cloud platforms including Amazon Bedrock and Google Vertex AI, with pricing unchanged from Opus 4.6.
- Delivers a 13% resolution lift on a 93-task coding benchmark, solving four tasks Opus 4.6 and Sonnet 4.6 could not.
- Features enhanced self-verification, better long-context performance, and higher-resolution vision for professional design tasks.
- Serves as a testbed for new cybersecurity safeguards, with limited capabilities to inform the safe future release of more powerful models.
Why It Matters
It enables developers to confidently offload their most difficult coding work, accelerating development velocity for complex, real-world applications.