Optiver Software Engineer Interview: The Full Process, Decoded
- The 80-in-8 mental math test eliminates more qualified engineers than the coding round — prep it first, two weeks minimum
- Beat the Odds tests expected value and betting decisions under time pressure, not just probability formulas
- Optiver's coding rounds are LeetCode medium difficulty, but narrating complexity and data structure tradeoffs matters as much as correctness
- C++ fluency is a meaningful advantage — production systems are C++, and interviewers notice candidates who can discuss memory layout and cache behavior
- "Why Optiver?" requires specific knowledge of market making or low-latency systems — generic answers are a red flag at a firm this specialized
- On-site system design focuses on low-latency market data distribution, not abstract web-scale architecture
- 8-10 weeks is a realistic timeline; mental math and probability require habit formation, not cramming
You've been grinding LeetCode for three months. You can explain a monotonic stack in your sleep. You have a clean sliding window solution and you know when to reach for a heap instead of a sorted array.
Then Optiver sends you the OA link. You open it. There are 80 arithmetic questions. You have 8 minutes. No calculator.
Good luck.
Optiver is a market maker. It runs on exchanges across equities, options, and ETFs, quoting buy and sell prices on thousands of instruments simultaneously. Its systems respond to market events in microseconds. The interview reflects that. You will face a mental math test, a probability game, cognitive reflex assessments, and only then a coding round. If you prep like it's a Google loop, you will not make it past the OA.
This guide walks through every stage in order, what each one actually tests, and how to prepare without wasting weeks on the wrong things.
Why This Interview Looks Nothing Like a FAANG Loop
Optiver is not a tech company that happens to do finance. It's a trading firm that needs engineers to build the fastest, most reliable execution infrastructure in the world. Order management systems. Pricing engines. Market data pipelines. Systems where a slow cache lookup costs real money in a very literal, millisecond-by-millisecond sense.
Software engineers here are not building web apps or internal tools. They are building execution platforms where a microsecond of latency is a product defect. The interview filters for people who think quantitatively under pressure and who understand that picking the wrong data structure isn't a code review comment. It's a production incident.
The Full Process, Since Nobody Warned You
| Stage | Format | Typical Length |
|---|---|---|
| Online Assessment | Mental math, probability, cognitive games, coding | 3-4 hours total |
| Recruiter Screen | Phone call, motivation, basic probability | 30-45 minutes |
| Technical Interview(s) | DSA coding, C++ / systems knowledge | 45-60 minutes each |
| On-Site Loop | Coding, system design, behavioral | 4-6 hours |
The OA is the biggest filter. Many candidates who would pass the technical rounds never see them because they underprepared for the math and probability sections.
Stage 1: The OA Nobody Warned You About
This is four separate sections, and each one is pass/fail. A great coding section does not compensate for a bad mental math score. They are independent hurdles.
80-in-8: How Fast Can You Do Long Division?
Eighty arithmetic questions in eight minutes. Addition, subtraction, multiplication, division, fractions, decimals. No calculator. Pen and paper are allowed, but at this pace you mostly won't use them.
Correct answers add a point. Wrong answers subtract a point. Unanswered questions score zero. If you're not confident on a question, skip it. The target to be competitive is around 70 correct.
The main failure mode is treating this like a trivia quiz. It's a speed test against your own nervous system. Your single-digit multiplication tables need to be cold, fraction simplification needs to be automatic, and two-digit multiplication shortcuts need to be in muscle memory before you sit down. Daily practice for two weeks, targeting 60+ questions in 8 minutes, is the minimum investment. Most engineers take three weeks to get comfortable.
The Optiver mental math test, right around question 30.
Beat the Odds: They're Testing If You Think Like a Trader
Thirty minutes. Ten questions, each with a 90-second cap. Each question presents a scenario like dice rolls, card draws, or sports outcomes, and asks you to calculate expected value and decide how to bet.
The questions look like: "You roll a fair six-sided die. If it shows a 6, you win $12. Otherwise you lose $2. What is the expected value of playing once?"
You're not just solving the math. You're also making allocation decisions under uncertainty. Optiver is testing whether you think like a market maker: quantify the edge, size the bet, don't anchor on intuition. You need to be comfortable with conditional probability, Bayes' theorem, and EV calculations before you sit this section.
Zap-N: Yes, There's a Balloon Game
Nine short games totaling one to two hours. They measure reflexes, attention, risk calibration, and pattern recognition. One game has you pumping up a balloon for increasing rewards until it pops and wipes your balance. Another has you removing food from a grill in a shrinking time window as more items are added.
These games are explicitly measuring behavioral traits: how you handle greed, how you calibrate risk, how fast you adapt to increasing complexity. The balloon game is essentially a greed detector with a UI. There is limited preparation here beyond getting a good night's sleep and doing a dry run if you can find sample games.
Coding Section: The Part You Actually Trained For
Two programming questions, two hours. One is a DSA problem (roughly LeetCode medium), one tests object-oriented design or implementation. Time complexity and correctness matter. Optiver has historically asked questions involving hash maps, priority queues, and simulation.
This is the easiest section relative to a standard FAANG OA. If you have solid LeetCode medium fluency, this is not your bottleneck. The mental math test is.
Stage 2: Recruiter Screen
A 30-45 minute phone call. The recruiter is assessing three things: your motivation for Optiver specifically, your communication, and basic quantitative reasoning.
"Why Optiver?" is not a throwaway question here. Generic answers about interesting problems or competitive compensation are a red flag at a firm this specialized. You need to demonstrate you understand what the firm actually does. Something grounded in market microstructure, low-latency systems, or Optiver's specific role in options markets will land far better than anything that could apply to any finance company or tech startup.
Expect one or two basic probability questions at this stage. Expected value, conditional probability, simple combinatorics. Nothing deeper than what appeared in the OA, but you should be able to answer verbally without writing anything down.
Stage 3: Technical Interviews: Where Narrating Beats Solving
After the recruiter screen, one or two technical interviews before the on-site. The format is typically a shared coding environment with a single engineer.
The coding questions are LeetCode medium in difficulty, but narrating your decisions, reasoning about complexity, and justifying your data structure choices is the actual job. A correct solution with no explanation gives the interviewer nothing to write. A 90% correct solution with clear reasoning about why a heap beats a sorted array here gives them plenty.
Optiver's interviewers care about several things beyond correctness:
- Can you reason about time and space complexity on demand?
- Do you know when a hash map is O(1) average and when that assumption breaks?
- Do you understand memory layout well enough to know that struct field ordering affects cache behavior?
- Do you understand pointer semantics and when a naive implementation becomes a performance liability?
C++ knowledge is a meaningful advantage. Optiver's production systems are written in C++. You don't need to be fluent to pass, but candidates who can discuss memory allocation, cache lines, and the cost of virtual dispatch are taken more seriously. If you're a Python or Java candidate, be ready to explain what your language does under the hood.
Stage 4: The On-Site Loop
Four to six hours of back-to-back interviews. The structure varies by team, but typically includes:
Coding rounds that go deeper than the earlier technical screens. You may be asked to implement a class from scratch, handle concurrency correctly, or extend your solution to handle ten times the load. Follow-up questions often push toward performance: "How would this behave with a million entries? What breaks first?"
System design with a low-latency emphasis. The prompts are not "design Twitter." They're closer to: design a system that aggregates market data from multiple exchange feeds and distributes it to trading algorithms with minimal latency. You should be able to discuss failover strategies, the tradeoff between UDP and TCP for market data, shared memory IPC, and why you would or would not use a database in a hot path.
Behavioral rounds that focus on collaboration under pressure, handling disagreement, and resilience after failures. Optiver moves fast. They want to know if you course-correct quickly or if you dig in on bad ideas. Concrete past examples beat abstract principles every time. "When I was wrong about the caching strategy on project X, here's what I actually did" is far better than "I would reassess and collaborate with the team."
The DSA That Actually Shows Up
The highest-value patterns to prepare:
- Hash maps and hash sets: frequency counting, window tracking, O(1) lookups
- Heaps: top-K problems, streaming median, merge K sorted lists
- Graphs: BFS/DFS on implicit graphs, topological sort, connected components
- Sliding window: subarray problems, window constraints
- Binary search: on answer space, not just sorted arrays
- Interval problems: merging, scheduling, overlap detection
Dynamic programming appears occasionally but is not as central as at Google or Meta. The real differentiator is whether you can articulate the complexity of what you wrote and identify the actual bottleneck.
For practice, SpaceComplexity runs voice-based mock interviews where you narrate your approach in real time. That format is closer to what Optiver's on-site actually feels like than silent LeetCode practice.
Mistakes That Kill Good Engineers' Chances
"Entry-level role." "Must pass 80-question mental math test in 8 minutes."
Skipping mental math prep. This is the most common way strong engineers fail the Optiver process. It doesn't feel like a real preparation task, so people skip it. Then they open the OA and wish they hadn't. Two weeks of daily practice closes most of the gap. One week is not enough.
Generic "Why Optiver?" answers. Interviewers at a trading firm notice when you could have given the same answer to Goldman Sachs or Bloomberg. Do the research. Know what a market maker actually does. Know what instruments Optiver trades. Know why sub-microsecond latency matters more here than almost anywhere else.
No probability preparation. Beat the Odds requires genuine expected value fluency. A textbook is overkill. Thirty EV problems under time pressure is not.
Treating the coding round as the main event. The OA has knocked out more qualified engineers than the technical interviews have. Sequence your prep to match the actual filters.
Going silent during on-site coding. Optiver's rubric rewards communication and reasoning. A candidate who solves every problem in silence is harder to evaluate than one who narrates confidently and misses an edge case. The interviewer is judging whether you would be a good collaborator in a fast-moving trading environment. Thinking out loud under pressure is a skill you can train separately from DSA.
How to Use 8-10 Weeks
Mental math and probability require habit formation, not cramming. You cannot speed-run 70 correct in 8 minutes over a long weekend.
- Weeks 1-2: Mental math daily (15-20 minutes), EV and probability problems
- Weeks 3-4: LeetCode mediums across the core patterns above (two per day, timed)
- Week 5: C++ or Java memory fundamentals, complexity deep dives
- Week 6: System design concepts for low-latency systems
- Weeks 7-8: Mock interviews, full OA practice under time pressure, behavioral prep
If your probability is weak, expand weeks 1-2. If you have a strong systems background, compress week 6. The mental math target is fixed regardless.
Key Takeaways
- The mental math test eliminates more qualified candidates than the coding round does. Prep it first.
- Low-latency, C++, and quantitative reasoning are first-class topics here, not tiebreakers.
- "Why Optiver?" requires real research. Generic answers are a red flag at a firm this specialized.
- DSA is LeetCode medium. Hash maps, heaps, graphs, sliding window, and binary search cover most of what appears.
- Narrate everything on-site. Reasoning out loud is part of the rubric, not a bonus.