Anthropic has released its original performance take-home assignment as an open-source repository. The challenge tasks engineers with optimizing code to beat Claude Opus 4.5's benchmark score of 1487 clock cycles. Candidates who succeed are encouraged to email their submissions for potential interviews.

This move offers a transparent look at the company's hiring process, which previously involved a timed test. By sharing the problem, Anthropic provides a public benchmark for AI coding performance and a practical tool for developers to hone their optimization skills against a known model's output.

The repository includes performance benchmarks for several Claude models, including Sonnet and Opus variants, measured in simulated clock cycles. It invites developers to attempt the challenge with unlimited time, contrasting the two-hour limit given to human candidates during the original hiring process. Success could lead to a recruiting conversation.