Anthropic has a 2-hour engineering take-home test. It says its new Claude 4.5 model outscored every human who took it.

Anthropic has unveiled its latest AI model, Claude Opus 4.5, claiming it has outperformed human candidates in a rigorous coding evaluation. The company announced on Monday that this new model achieved higher scores than any human who has ever taken their notoriously challenging two-hour take-home exam designed for prospective engineering candidates. In a detailed blog post, Anthropic explained that the test evaluates candidates' technical skills and decision-making abilities under pressure. While the test does not encompass all necessary engineering competencies, the results raise important questions about the future of engineering professions in light of AI advancements. The model's success was attributed to its ability to work through multiple attempts at solving each problem, ultimately selecting the best response. While specifics about the test format remain largely undisclosed, a 2024 review on Glassdoor indicated that it consists of four stages where candidates must implement and enhance a specific system. It is not confirmed whether Claude 4.5 faced the same version of the test. This latest iteration comes only three months after the previous Claude model and introduces enhancements beyond coding, including improved capabilities for generating professional documents like Excel files and PowerPoint presentations. Anthropic appears to be solidifying its position in the AI coding landscape, with even competitors like Meta leveraging Claude for their internal coding assistant, Devmate. The company's training methodologies remain closely guarded. Eric Simons, CEO of Stackblitz, speculated that Anthropic's approach may involve having its AI generate and execute code independently, subsequently reviewed by human experts and AI tools. Dianne Penn, Anthropic's Head of Product Management, confirmed that this depiction aligns closely with their practices. At the recent Dreamforce conference, Anthropic CEO Dario Amodei remarked that Claude AI is currently responsible for producing approximately 90% of the code written by teams within the company. However, he emphasized that this does not equate to a reduction in software engineering roles. Instead, he suggested that the presence of AI allows engineers to concentrate on the more complex aspects of coding and oversight, potentially increasing overall demand for human engineers.