The Anthropic Online Assessment (CodeSignal)

The Anthropic online assessment is a CodeSignal "industry coding framework" build: one stateful problem that grows across four locked levels, auto-graded out of 600, with about ninety minutes on the clock. The reported bar is roughly 520 of 600 points, clearing at least three of the four levels, and each level unlocks only after every hidden test for the level below it passes. The unit tests are visible before you write a line. It is the technical screen for most Anthropic engineering pipelines, and recruiters describe it up front as practical, not LeetCode. They mean it literally.

Level one is usually a small working system, a common one being an in-memory key-value store with SET, GET, and DELETE. Level four bolts new behavior onto that same store, timestamped writes, time-to-live expiry, file compaction and a capacity limit, and the methods you wrote at level one still have to pass. The whole assessment works like that. You build a small system, the requirements keep arriving, and the code that survives is the code you can extend without tearing it down. A version that compiles and clears the first case is the warm-up. The grade lives in everything after it.

The Anthropic angle

The company that builds Claude tells you not to use Claude on the assessment. Anthropic's candidate guidance is blunt: this is all you, no AI assistance unless an interviewer indicates otherwise, and the same rule covers take-homes. Anthropic also reportedly screens for code tuned to clear the visible tests rather than genuinely solve the problem. That is part of why the level-one methods still have to work after level four adds behavior: test-gaming shortcuts tend to break the later levels. Confirm the policy with your recruiter, then plan to work from memory in a shared editor.