Anthropic has repeatedly revised its take-home technical interview test for job applicants since 2024 to mitigate AI-assisted cheating. The performance optimization team, responsible for administering the test, found that advancements in AI coding tools necessitated these changes.
Team lead Tristan Hume stated in a Wednesday blog post that each new Claude model has compelled test redesigns. Claude Opus 4 surpassed most human applicants when given the same time limit, while Claude Opus 4.5 matched the performance of top candidates. This eliminated Anthropic’s ability to differentiate between the work of leading human applicants and its most advanced AI model under the take-home test conditions.
Hume developed a new test focusing less on hardware optimization, making it sufficiently complex to challenge current AI tools. The original test was also shared in the blog post, inviting readers to propose alternative solutions. The post indicated, “If you can best Opus 4.5, we’d love to hear from you.”
