Anthropic has revealed that its latest AI model, Claude Opus 4.5, outperformed every human candidate who took its notoriously challenging two-hour engineering take-home test. The company described the new model as its most advanced yet and highlighted the achievement as a milestone in AI’s growing role in software development. In a blog post, Anthropic stated that Claude Opus 4.5 scored higher than any human applicant on the coding assessment, which is designed to evaluate technical problem-solving skills, coding proficiency, and decision-making under time pressure. While the test doesn’t capture all aspects of an engineer’s capabilities—such as collaboration or system design—it does represent a significant benchmark for technical aptitude. The company explained that the model was given multiple attempts to solve each problem, with the best result selected for evaluation. This approach reflects how Anthropic tests its models: by allowing them to iterate and refine their responses, mimicking the way human engineers might debug and improve code. Details about the test’s exact content remain limited. A 2024 Glassdoor interview review described the assessment as having four levels, requiring candidates to build and extend a system with specific functionalities. It is unclear whether the version used to test Claude Opus 4.5 matched this structure, as Anthropic did not provide further specifics in its announcement and declined to comment on request. The release of Claude Opus 4.5 comes just three months after the launch of its predecessor, underscoring the rapid pace of development at Anthropic. Beyond coding, the new model also features enhanced capabilities in generating professional documents, including Excel spreadsheets and PowerPoint presentations. This latest achievement reinforces Anthropic’s leadership in AI-powered coding. Even Meta, a direct competitor in the AI space, has integrated Claude into its internal Devmate coding assistant, highlighting the model’s practical value despite the broader rivalry. Anthropic has not disclosed its training methodology, but Eric Simons, CEO of Stackblitz and founder of the coding platform Bolt.new, previously told Business Insider that he believes Anthropic’s AI models write and deploy code autonomously, with human and AI reviews following. Dianne Penn, Head of Product Management, Research and Frontiers at Anthropic, confirmed that this description was “generally true.” At the Dreamforce conference in October, Anthropic CEO Dario Amodei revealed that Claude is already responsible for writing 90% of the code used by most teams within the company. However, he emphasized that this does not mean replacing human engineers. Instead, he said, AI allows developers to focus on the most complex, creative, and strategic parts of software development—the 10% that require human judgment, oversight, and innovation. “You might need more engineers, not fewer,” Amodei said, “because they can now be more productive and leverage AI to handle the bulk of routine work.”