People apply to “Citadel” every year without registering that they’ve applied to one of two companies. Same founder, same logo, often the same tower in Miami, and two genuinely different jobs sitting behind two different interview loops. Mixing them up is the quickest way to walk into a room prepared for the wrong thing.
Ken Griffin runs both. Citadel is the multistrategy hedge fund he started in 1990. Citadel Securities is the market maker he spun up in 2002, and it now prices a large slice of US retail equity order flow. They share a brand and a reputation for being hard to get into, but they are separate legal entities with separate teams, separate comp bands, and separate hiring timelines. You can interview at one, get dinged, and still apply to the other.
Why the distinction changes how you prep
The hedge fund makes money by taking positions across equities, fixed income, commodities, and credit, and holding them over horizons measured in days to months. Research there bends toward signal generation, portfolio construction, and statistics that survive out of sample. Citadel Securities makes money on the spread, thousands of times a second, so its center of gravity sits closer to market microstructure, execution, and the systems that quote and hedge in microseconds.
| Citadel (hedge fund) | Citadel Securities (market maker) | |
|---|---|---|
| Core business | Multistrategy investing | Electronic market making |
| QR emphasis | Alpha signals, statistics, portfolio risk | Microstructure, execution, short-horizon prediction |
| Engineering emphasis | Research platforms, data, risk systems | Low-latency C++, exchange connectivity |
| Trading role | Systematic and discretionary analysts | Market-making and execution traders |
| Holding period | Days to months | Microseconds to a day |
The probability and coding bar is similar at both. What shifts is the flavor of the applied questions and which desk wants you once the technical rounds clear.
The loop, end to end
Plan for four steps over roughly four to five weeks. A recruiter screen comes first, usually 30 minutes on your background and what you actually want to do. Then an online assessment, then one or two technical phone rounds, then a final round (the superday) of three to five back-to-back interviews. After you’ve cleared the technical bar, you go through team matching, where hiring managers decide who wants you. Strong candidates sometimes get pulled by more than one desk and get to choose.
If you’re a student, the side door matters. Citadel runs recruiting competitions like the Datathon and the Data Open, plus Terminal, a programming game where bots fight on a grid, and it pulls interns and full-timers straight out of those. Placing well there can skip you past the resume pile entirely.
Quantitative Researcher: probability until it hurts
The QR online assessment is where most people get cut, and it leans on probability and statistics, not LeetCode grinding. Expect timed questions on expected value, conditional probability, and distributions, with a few that reward spotting a clean trick over brute force.
A question candidates report almost verbatim: two people each arrive at a station at a uniformly random time between 4 and 5pm, and each waits ten minutes for the other. What’s the chance they meet? Draw the unit square, subtract the two corner triangles, and you get 1 – (50/60)2 = 11/36, about 0.31. The interviewer cares less about the number than whether you reached for the geometric picture instead of flailing at an integral.
Some others that show up in one form or another:
- Expected number of fair-coin flips to see two heads in a row (it’s 6, and they want the recurrence, not the memory).
- A stick breaks at two uniformly random points; what’s the probability the three pieces form a triangle?
- A test is 99% accurate and the disease hits 1 in 10,000. You test positive. Now reason about how worried you should be.
- Ridge versus lasso: what does each do to the coefficients, and when would you pick one over the other?
The statistics half goes deeper than the assessment. Linear regression assumptions, PCA, and overfitting come up constantly, usually wrapped in a “here’s a modeling situation, what would you do” frame rather than a definition quiz. Say the quiet part out loud: name your assumptions, say what would break them, say how you’d check. They are watching how you handle uncertainty, which is the actual job.
Quant Developer and Software Engineer: C++ and where the latency hides
Quant Developers live in C++. The interview probes memory layout, cache behavior, move semantics, low-contention data structures, and the gap between code that is correct and code that is fast on real hardware. A common move is a normal algorithm question followed by “now make this run in a tight loop a million times a second without allocating.”
The Software Engineer loop is closer to a hard FAANG bar with a finance accent: data structures and algorithms in a CoderPad session (Python or C++ are both fine), then system design. Design prompts skew toward data and reliability, like a market-data distribution layer, an order gateway, or a time-series store that has to survive a bad tick without poisoning everything downstream.
- Given a stream of trades, keep a rolling median updated in O(log n) per tick.
- Build a thread-safe queue for a single producer and single consumer with minimal contention.
- Why might a hash map be slower than a flat array of structs for hot-path lookups?
- Design the path from an exchange feed to a strategy that needs the order book current within microseconds.
Quantitative Trader: making a market out loud
Trader interviews swap proofs for speed. You’ll do mental math against a clock (multiply 47 by 63, go), and you’ll play market-making games where the interviewer asks you to quote a two-sided market on something with an unknown value, then trades against you and updates the situation. The skill is sizing your spread to your uncertainty and adjusting as information arrives, not nailing the first guess.
One classic: “There’s a jar of jellybeans on the table; make me a market.” You quote a bid and an ask, they hit or lift one side, and you learn from which side they took. Stay coherent, update fast, and don’t freeze when you find out you were wrong.
What the offers actually look like
Both firms pay at or near the top of the market, and the numbers move enough that you should treat any figure as a snapshot. New-grad software engineering total compensation tends to land in the low-to-mid six figures once base, sign-on, and bonus are stacked. New-grad quant researcher packages run higher, often well into the mid six figures all-in at the top end, because the pool of people who clear that probability bar is thin. Internships pay at rates that, annualized, embarrass most full-time offers elsewhere.
Rather than trust a number in a blog post, pull current data from levels.fyi and recent Glassdoor and Wall Street Oasis entries, and weight reports from the last year over anything older. Pay at this tier is bonus-heavy and swings with how the fund did and which team you land on, so the headline base is the least interesting part of the package.
How to spend the month before
If you’re aiming at research, drill probability and expected value until the standard setups (geometric probability, Markov chains, Bayes, gambler’s ruin) are reflexes, and be ready to defend a regression you’ve actually built rather than one you read about. For developer and engineering roles, get fast and clean in C++ or Python, then practice talking through latency and memory instead of just passing the test cases. For trading, do timed mental math every day and play market-making games with a friend who will actually trade against you and try to pick you off.
What separates offers from rejections here is rarely raw cleverness. It’s whether you can hold a hard problem in your head, narrate your reasoning so the interviewer can follow it, and stay steady when they push back or quietly change the question. Citadel and Citadel Securities both hire for that, in slightly different keys.
