The quant brainteaser is not one test. It is several different tests wearing the same name, and preparing for all of them the same way wastes the few weeks you have. A Jane Street phone screen, an Optiver online assessment and an SIG onsite each filter for a different sub-skill, and a 2026/27 Optiver invite may not even contain the test the prep sites still describe.
Across the market makers, the first gate is speed, not depth. techinterview.org's 2026 quant-finance guide (updated 12 July 2026) groups Jane Street, Optiver, SIG and "most market makers" as mental-math-first, and says that below roughly 80% on the timed drill "you're out regardless of other strengths." The multi-strategy and systematic shops — Citadel, Two Sigma — open with a probability-and-statistics screen too, but their questions reward derivation over arithmetic. Work out which door you are walking through before you drill, because the prep is not interchangeable.
What each firm actually screens for
Every figure below is third-party or candidate-reported; none is published by the firm. Dates are when the source was last updated.
| Firm | First gate | Signature question style | What it really tests |
|---|---|---|---|
| Jane Street | ~60-question, 8-minute mental-math test (techinterview, 16 Oct 2026) | Expected value of betting games, random walks, card and dice puzzles | Structured reasoning out loud, betting under incomplete information |
| Optiver | 80-in-8 arithmetic, or its 2026/27 replacement modules (Quantt, 29 Sep 2026) | Fast fair-value and probability against a clock | Raw speed and accuracy first, intuition second |
| SIG (Susquehanna) | Timed probability/brainteaser assessment, "less mental-math-heavy than Optiver's 80-in-8" (techinterview, 21 Jul 2026) | Pot-odds and decision-theory problems; poker optional | Expected-value decisions and discipline under variance |
| Citadel | Probability-and-stats screen, then a coding round (Aced, Oct 2026) | Bayes scenarios, evaluate-a-strategy statistics | Judgment and clean code, not puzzle trivia |
| Two Sigma | HackerRank with Python and linear regression (Aced) | OLS derivations, hypothesis testing, modeling cases | Statistics depth; the guide says not to specialize |
Count the pattern: the two market makers at the top screen for how fast and clearly you move; the two funds at the bottom screen for whether your statistics are real. SIG sits in between, testing the quality of your decisions.
The mental-math gate is its own exam
The timed arithmetic test is not a probability test, and treating it as one is the first mistake. It is pure speed.
techinterview.org's Jane Street guide (updated 16 October 2026) describes about 60 questions in 8 minutes — two-digit by two-digit multiplication, division, percentages, decimals and fraction comparison — with no calculator and a pass bar it gives as "roughly 70-80%." The guide flags its own figure as an unsourced estimate and notes it lists the bar two ways, so take the exact number as approximate. The recommended drill is Zetamac, and the standard the guide quotes (60+ correct in two minutes of practice) is far harder than the real test, by design.
Optiver's version is the famous 80-in-8. Quantt's Optiver guide (published 9 April 2026, updated 29 September 2026) reports it as 80 four-option multiple-choice arithmetic questions in 8 minutes — about six seconds each — scored +1 for correct and -1 for wrong, with the most repeated pass mark around 55-56 net and 70-plus "commonly" competitive. The negative marking is the trap: blind guessing has zero expected value, so you answer only what you can actually compute in the time.
The important update is that the test may be gone. Quantt says the modules 2026/27 invitees are quoting list Beat the Odds, NumberLogic (about 26 questions in 25 minutes, calculator allowed) and new Likelihood-list, Intervals and Orderbooks sections — with no 80-in-8 — and that a prep site claiming it is still in use is unconfirmed. One official rule survives either way: Optiver's Europe FAQ, cited by Quantt, says its assessments "can only be completed once every 8 months," so you do not get to burn a sitting as a practice run.
SIG's assessment leans the other way. techinterview.org's SIG guide (updated 21 July 2026) calls the online test "timed and demanding" but "less mental-math-heavy than Optiver's 80-in-8," weighting probability, brainteaser and quantitative-reasoning questions over raw arithmetic. And the software-engineering track at Jane Street skips the drill entirely — its SWE interview guide says those loops do not include mental math, math-olympiad questions or logic puzzles, so candidates who grind Zetamac for a software role are preparing for the wrong interview.
Expected value, not the most likely outcome
Once you are past the timer, almost every firm is testing the same instinct: price the bet, do not predict the future. This is clearest at SIG, where many senior traders are former competitive poker players.
The SIG guide's sample is a pot-odds question — you have a 30% chance of winning a hand, the pot is $100, and your opponent bets $50 — and the point is not the fold-or-call answer but that you reason in expected value out loud. Poker is not required; the guide says non-players get "a structured decision-theory problem" in plain language that uses the same logic. The onsite can still include an actual poker hand, and what it measures is bet sizing, updating on new information, and how you behave after a bad beat.
Jane Street frames the same instinct as betting games. Its guide lists a dice game against one opponent, a take-turns stone-pulling game where the last mover wins (optimal strategy asked), and a die-acceptance game asking for the right threshold. The classic warm-up it names — "I flip a coin until I get two heads in a row; what's the expected number of flips?" — has the tidy answer of 6 for a fair coin, and the quant-finance guide's market-making twist is to follow it with "what would you bid or offer on this?" You are expected to quote a two-sided market and tighten the spread as your uncertainty falls.
The question types, by how they are graded
| Type | Example (with source) | Firms that lean on it | How it is scored |
|---|---|---|---|
| Expected-value games | Expected flips for two heads in a row; expected rolls to the first 6 (techinterview; Aced) | Jane Street, SIG, D. E. Shaw | Setup and EV reasoning, read aloud |
| Bayesian updating | Citadel intern "Bayes' theorem scenario, then an open brainteaser" (Aced) | Citadel, SIG, Point72/Cubist | Correct conditioning, not memorized formula |
| Combinatorics and cards | Jane Street "deck numbered 1-5, which composition minimizes expected value" (Glassdoor) | Jane Street, Citadel | Clean counting under time |
| Statistics and OLS | Two Sigma "OLS proof via Lagrange multipliers"; t-test sample size (Aced; Glassdoor) | Two Sigma, AQR, Citadel | Derivation depth, estimator properties |
| Market-making games | "Make me a market on the number of dominoes in a set" (techinterview SIG) | Optiver, SIG, Jane Street | Fermi anchor, spread discipline, inventory |
The split in that table is the whole strategy. For Jane Street and SIG, the grade is your spoken reasoning on a bet; silent computation of a correct number can still cost you. For Citadel and Two Sigma, the grade is whether your probability and statistics hold up under follow-ups — Aced reports a Citadel statistics question on how you would evaluate a trading strategy, and Two Sigma loops of up to seven rounds (three technical, three with hiring managers, one with an MD) spanning probability, statistics, ML, SQL and Python, with the explicit advice not to specialize.
What gets people rejected
1. Drilling arithmetic for a job that tests statistics. Zetamac buys you nothing at Two Sigma's OLS-derivation round. Match the prep to the firm's first gate in the table above.
2. Going silent on a betting problem. The Jane Street and SIG guides both say interviewers vary the problem to catch rote answers and grade the reasoning, so narrate it.
3. Guessing through negative marking. On Optiver's +1/-1 scoring, guessing is expected-value-zero. Skip what you cannot compute in six seconds.
4. Preparing for a test that has been retired. Quantt's 29 September 2026 update says recent Optiver invites may not contain the 80-in-8 at all. Read your actual invite before you build a plan around a prep-site format.
5. Burning your one sitting. Optiver's 8-month cool-off means a careless attempt locks you out for most of a recruiting cycle.
6. Preparing for the wrong loop. Jane Street's software-engineering interview has no mental-math or puzzle rounds; its quant interview is built on them. Confirm the role.
A 3-week plan
- Week 1: map the gate, then drill speed. For each target, copy the first-round format from the table and the Sources below, and confirm it against your own invite. If a market maker is on your list, do daily Zetamac sets; if a fund is, start a probability-and-statistics problem set instead. Track which firms have opened on the New Grad 2027 tracker.
- Week 2: expected value out loud. Work twenty mixed EV and Bayesian problems a day — coin and dice stopping times, conditional-probability puzzles, pot-odds decisions — and solve every one by talking through the setup, not just writing the answer. Add one make-a-market Fermi estimate (gas stations, dominoes) each day and practice quoting a two-sided spread.
- Week 3: mock the real rounds. Run full sessions against the format you will face. Applr's Jane Street quantitative researcher mock runs the back-to-back probability format; the SIG quantitative trader mock pushes pot-odds and decision-theory questions; and the Citadel quantitative researcher mock mixes Bayes and evaluate-a-strategy statistics. Still choosing a coding language for the engineering rounds? Our C++ vs Python guide reads the 2027 postings firm by firm.
For international students
The timed first gate is the one stage where a visa, a time zone and an accent cost you nothing — it is scored on speed and correctness alone, which is why it is the best place to build an edge if you are applying from outside the firm's home market. Two things still need planning. First, the EV and poker rounds are judged on how you reason aloud, so rehearse explaining probability in English, not just solving it; the quant-finance guide says interviewers grade whether you "communicate concisely." Second, when the loop moves to in-person onsites, a candidate studying far from a New York or London office needs to book travel early, and should ask the recruiter whether finals can run virtually. On seats, confirm sponsorship on the specific posting before you spend a sitting you cannot retake for eight months.
Sources
- techinterview.org, "Jane Street Interview Guide 2026", updated 16 Oct 2026
- techinterview.org, "SIG (Susquehanna) Interview Guide: Poker, Options & Decision Theory", updated 21 Jul 2026
- techinterview.org, "Breaking Into Quant Finance and Wall Street: 2026 Guide", updated 12 Jul 2026
- techinterview.org, "What Jane Street's software engineer interview really tests", 2026
- Quantt, "Optiver Interview: Process, Questions", published 9 Apr 2026, updated 29 Sep 2026
- Aced, "Quant Researcher Interview Guide", 2026
- Tradermath, "Quant Interview Questions", 17 Jan 2026
- Quantt, "Quant Interview Questions", 2026
- 1Point3Acres, Optiver 2026 Quantitative Research Internship online assessment report (candidate-submitted)
Frequently asked questions
- What probability questions do quant interviews actually ask?
- Expected-value problems dominate: expected number of coin flips to get two heads in a row, expected rolls of a die to see the first 6, the expected position of the first ace in a shuffled deck, and random walks that ask the chance of hitting +N before -M. These examples are listed in techinterview.org's Jane Street guide (updated 16 October 2026) and Aced's quant researcher guide. Firms care more about the reasoning out loud than the final number.
- Is Jane Street's mental math test really 60 questions in 8 minutes?
- That is the figure techinterview.org's 2026 guide gives (about 60 questions in 8 minutes, no calculator, pass bar roughly 70-80%), but it is a third-party estimate, not a number Jane Street publishes. The guide also notes variants exist and lists the pass bar two ways. Treat it as an order-of-magnitude target, not a rule.
- What is Optiver's 80-in-8 test?
- 80 multiple-choice arithmetic questions in 8 minutes, scored +1 for a correct answer and -1 for a wrong one, with a candidate-reported pass mark around 55-56 net (Quantt, updated 29 September 2026). Quantt adds that 2026/27 invites candidates quote list newer modules (Beat the Odds, NumberLogic and others) with no 80-in-8, so confirm what your invite actually contains.
- Do you need to play poker to interview at SIG?
- No. techinterview.org's SIG guide (updated 21 July 2026) says poker is not required and candidates who don't play get a decision-theory problem in plain language instead. What SIG screens for is expected-value reasoning and discipline under variance; a sample pot-odds question asks whether to call a $50 bet into a $100 pot with a 30% chance of winning.
- How are Citadel and Two Sigma quant interviews different from market makers?
- Both put a probability-and-statistics screen first, but reward derivation over arithmetic speed. Aced's guide reports Citadel intern rounds opening with a Bayes' theorem scenario and an evaluate-a-trading-strategy statistics question, and Two Sigma loops of up to seven rounds covering OLS derivations, hypothesis testing and modeling cases, with the advice not to specialize.