Anthropic's Latest Claude Model Delivers for Developers, But Its Best AI Remains Under Lock and Key

Anthropic has released Claude Opus 4.7, a model that gives software engineers a significant upgrade for complex, hands-off coding tasks. Available now, it outperforms key rivals on standard benchmarks. On the SWE-bench Pro test, it scored 64.3%, ahead of OpenAI's GPT-5.4 (57.7%) and Google's Gemini 3.1 Pro (54.2%).

Yet, this public release is not Anthropic's top performer. That title belongs to a restricted system called Mythos Preview, available only to select partners for security testing. Mythos scores dramatically higher, with Anthropic stating its advanced capabilities in areas like finding software vulnerabilities pose a safety risk if widely released. Opus 4.7 includes safeguards against such high-risk requests.

For working developers, the update brings concrete improvements. Performance on coding benchmarks like CursorBench jumped from 58% to 70%. The model makes fewer errors when using tools, better handles multi-step projects, and can coordinate parallel tasks like code review and documentation. It also processes higher-resolution images, useful for translating UI mockups into code.

Pricing remains unchanged, and the model is accessible through Anthropic's API and major cloud platforms. The release continues a pattern of rapid iteration from Anthropic, following the Opus 4.6 model from February.

Early user reports highlight a reduced need for manual oversight, with the model more strictly following instructions and verifying its own work. However, comparisons show it still trails behind OpenAI's model in web search tasks and, of course, its locked-down sibling Mythos in raw performance. For businesses and developers today, Opus 4.7 represents the practical, available edge in automated coding, while the frontier of what's possible remains in a closed lab.

Source: Webpronews