Point72 Online Assessment: HackerRank + Cubist Take-Home

Point72 is Steve Cohen's multi-strategy hedge fund, headquartered in Stamford, Connecticut, running fundamental long/short pods alongside Cubist Systematic Strategies, its dedicated quant arm. Because those two sides hire very differently, there is no single "Point72 OA" — the online assessment you get depends heavily on which role and which entity you applied to. This page covers the screening stage specifically; for the interview rounds that follow, see our Point72 interview questions guide.

The reported OA formats, by role

Candidate reports on Glassdoor, Wall Street Oasis describe several distinct formats as of mid-2026. Treat the exact question counts as indicative rather than guaranteed — firms adjust these between cycles.

What the HackerRank actually tests

Reports of the developer-flavored test consistently describe medium-to-hard algorithm problems wrapped in business scenarios: you translate a requirement into a model, then care about complexity. Recurring themes candidates mention are greedy algorithms, efficient array traversal and grouping, and light number theory. For quant-titled roles, the mix shifts toward probability and statistics questions alongside shorter coding tasks — think conditional probability, expectations, and distribution reasoning under time pressure rather than pure LeetCode. Some candidates also report a separate short personality-plus-quick-math component and a recorded video round before any human interview, so budget for two or three online stages, not one.

This structure is typical of the pod-shop cluster: the screens at Citadel and Millennium follow a similar HackerRank-first pattern, and candidates frequently recycle preparation across all three.

The Cubist take-home

Cubist, Point72's systematic arm, hires pod by pod, and candidates report that each pod may run its own take-home modeling exercise instead of (or in addition to) a timed OA. The commonly described shape is an open-ended dataset exercise: build a predictive model or extract a signal, document your choices, and defend them in the 5–6 technical interviews that reportedly follow. That means the evaluation is less about a clever trick and more about research hygiene — sensible cross-validation, honest out-of-sample reporting, and no data leakage. Our machine learning question bank covers exactly the overfitting and validation traps interviewers probe when they dig into your write-up. Point72's own recruiting blog echoes this: they want candidates who can explain their research to someone outside their field.

What comes after the OA

Candidates who clear the online stages report phone or video technical screens, then a final round; one commonly cited format is a superday of roughly three 30-minute interviews, though this varies by team. The end-to-end process is frequently described as long — several weeks to a few months — especially for Cubist, where per-pod matching adds calendar time. Expect the live rounds to re-test the same material as the OA at a deeper level: probability brainteasers, statistics and inference questions, and code review of your take-home.

How to prepare

• Timed coding reps. Medium-difficulty greedy and array problems under a clock are the single highest-yield preparation for the HackerRank stage. Practice finishing 3 problems in 90 minutes, not just solving them eventually.
• Probability and stats under time pressure. The quant-role screens reward fast, correct expectation and conditional-probability calculations more than exotic theory.
• SQL basics. A minority of reports mention SQL questions; joins, aggregation, and window functions cover what's described.
• For Cubist: one clean end-to-end project. Rehearse the full loop — explore, feature-engineer, validate, write up — on a public dataset so the take-home format feels routine.
• Don't skip the soft screens. The personality and video stages filter real candidates; answer consistently and concretely rather than gaming them.

Ready to drill? Work through our probability question bank and coding interview questions under timed conditions, or browse the full problem bank — 2,800+ problems with worked solutions, around 400 of them free.