Goldman Sachs Software Engineer Interview (2026)

TL;DR: Goldman Sachs runs software engineers through four stages: an online assessment (HackerRank coding, sometimes with math multiple choice), a HireVue video, one or two live CoderPad screens, and a Superday of three to five interviews. What separates Goldman from a standard FAANG loop is the quant layer. Budget roughly 40 percent of your prep on core data structures and algorithms, 20 percent on probability and math, and the rest on behavioral and light system design. The whole process tends to run several weeks.

We run mock interviews with engineers targeting banks and trading firms every day, and the pattern is consistent: candidates arrive drilled on arrays and trees but freeze on a one-line expected-value question. At Goldman, that gap is expensive.

The process at a glance

Goldman does not use one fixed loop. Campus and new-grad candidates usually hit a HireVue early, while experienced engineers often start with a recruiter call and go straight to a coding screen. The stages below reflect what candidates most commonly report.

Stage	Format	What is evaluated
Resume screen	Recruiter review	Skills, education, relevant experience
HireVue video	~5 recorded questions, ~2 min each	Communication, motivation
Online assessment	HackerRank, 2 to 3 problems, ~60 to 120 min	Data structures and algorithms
Technical screen	Live CoderPad, ~45 to 75 min	Working code, testing, explanation
Superday	3 to 5 interviews, 45 to 60 min each	Coding, system design, behavior

For timeline, Interview Kickstart, reading Glassdoor data, puts the average at roughly 54 days end to end, with five to six rounds overall. Treat that as a rough average, not a promise; experienced hires with an urgent team can move faster.

Stage by stage

Online assessment and HireVue

The front door for most tech roles is a HackerRank online assessment. Candidate reports and prep guides converge on two to three coding problems in roughly 60 to 120 minutes, auto-graded on hidden test cases, though some proctored versions run longer and mix in multiple-choice fundamentals. For strats and quant-leaning roles, expect the assessment to bolt on probability and math multiple choice, so a coding-only prep plan leaves points on the table.

The HireVue is a recorded video interview with no live interviewer. You see a question, get around 30 seconds to think, then record an answer of up to two minutes. The questions are behavioral and motivational: why Goldman, why engineering, a time you handled a tradeoff. Record a few practice answers and watch them back. Most people ramble; the fix is a tight structure, and the STAR method is the cleanest one for this format.

Live technical screen

Clear the assessment and you get one or two live screens with an engineer, usually in a shared editor like CoderPad, part coding and part walking through something on your resume. Goldman interviewers care more about your reasoning than a memorized optimal solution. The recurring theme from candidate reports is to explain the engineering decision, not just the final answer: why this data structure, what the complexity is, which edge cases break it, and how you would test it.

The Superday

The final round is the Superday, typically three to five back-to-back interviews of 45 to 60 minutes. Expect a mix: more coding, a hiring-manager conversation, behavioral depth, and for some teams an open-ended design discussion. One reported design prompt involved handling high-volume data uploads. Treat specific prompts as examples, not a fixed script, and practice turning a vague requirement into a structured design with stated tradeoffs.

The quant and math layer

This is the part candidates underprepare, and it is where Goldman differs most from a generic tech loop. Even on pure engineering tracks, probability and mental-math questions show up, and on strats they are central. You do not need measure theory. You need fast, correct reasoning on expected value, conditional probability, and combinatorics. Two worked examples in the style Goldman favors:

Example 1, expected value with a decision. You roll a fair six-sided die once. You may keep the result or re-roll exactly once and take the second roll. Playing optimally, what is the expected value?

Re-roll only when the first roll is below a fresh roll's EV (3.5),
so re-roll on 1, 2, 3 and keep 4, 5, 6.

E = P(keep) * mean(4,5,6) + P(re-roll) * 3.5
  = (1/2)(5) + (1/2)(3.5)
  = 2.5 + 1.75
  = 4.25

Example 2, conditional probability. A family has two children. You learn at least one is a boy. What is the probability both are boys?

Equally likely ordered outcomes: BB, BG, GB, GG.
Condition on "at least one boy" removes GG, leaving {BB, BG, GB}.
Only BB has two boys.
P(both boys | at least one boy) = 1/3  (not 1/2)

If the 1/3 answer surprises you, that is exactly the trap these questions are built around. Drill enough of them that conditioning becomes reflexive. Khan Academy's statistics and probability track covers the fundamentals for free, and "A Practical Guide to Quantitative Finance Interviews" by Xinfeng Zhou (widely known as the green book) and "Heard on the Street" by Timothy Falcon Crack are the standard problem banks for this style.

Coding problems to drill

Across DevInterview's Goldman Sachs question bank, the most frequently reported problems skew toward strings, arrays, linked lists, and a few harder binary-search and DP questions. Map each to its pattern rather than memorizing solutions:

Problem	Level	Pattern
Two Sum	junior	Hash map lookup
Add Two Numbers	mid/senior	Linked list, carry handling
Longest Substring Without Repeating Characters	mid/senior	Sliding window
Container With Most Water	mid/senior	Two pointers
Longest Palindromic Substring	mid/senior	Expand around center / DP
String to Integer (atoi)	mid/senior	String edge cases
Integer to Roman	mid/senior	Greedy / math mapping
Median of Two Sorted Arrays	staff	Binary search on partitions

A few judgment calls from watching these play out in mocks. "String to Integer (atoi)" looks trivial and sinks people on edge cases: leading whitespace, sign handling, overflow clamping to the 32-bit range. Slow down and enumerate those before you type. "Median of Two Sorted Arrays" is the one staff-level problem here; if you are interviewing above senior, know the O(log(m+n)) partition approach cold rather than the merge shortcut. "Container With Most Water" is a clean two pointers warmup that interviewers love because the brute force is obvious and the optimization is a genuine insight. You can drill the full set with AI feedback on our practice page.

A 4-week prep plan

This assumes you are already working and prepping part time. Adjust the ratio toward quant if you are targeting strats.

Week 1: Rebuild fundamentals. Arrays, hashing, strings, linked lists, two pointers, sliding window. Solve "Two Sum", "Add Two Numbers", "Longest Substring Without Repeating Characters", and "Container With Most Water" untimed, then re-solve timed.
Week 2: Add binary search and DP ("Median of Two Sorted Arrays", "Longest Palindromic Substring") plus the string parsing of "Integer to Roman" and "String to Integer (atoi)". Start the math layer: one probability set per day. Begin mock interviews, 3 sessions of 45 minutes this week.
Week 3: Full loop simulation. Keep 3x 45-minute mocks per week, alternating coding and behavioral. Add light system design: practice one open-ended prompt (scalable file uploads, a rate limiter) end to end.
Week 4: Taper and polish. Record HireVue-style answers and watch them back. Re-solve any problem you stumbled on. Keep a daily 15-minute mental-math warmup so expected-value questions stay reflexive.

For the math grind, use "Heard on the Street" or the green book for problems and Khan Academy to patch gaps. For coding, HackerRank mirrors the assessment environment, and a concept-by-concept drill on math and the broader Goldman Sachs breakdown keeps you honest about coverage.

FAQ

How hard is the Goldman Sachs coding assessment compared to FAANG?

The coding difficulty is comparable to a mid-level FAANG screen: mostly LeetCode easy to medium with an occasional hard. The difference is breadth. Goldman mixes in probability, math, and sometimes multiple-choice fundamentals, so raw algorithm speed is not enough on its own.

Does every candidate get a HireVue and a HackerRank test?

No. Routes vary. Campus and new-grad applicants usually record a HireVue early, while experienced engineers often skip it and go from a recruiter call straight to a live coding screen. Some candidates see both; some see neither.

How much probability do software engineers really need?

For pure engineering roles, enough to handle an expected-value or conditional-probability question confidently if it comes up. For strats and quant-dev roles it is central and you should prepare it as seriously as coding. Either way, the fundamentals (EV, conditional probability, combinatorics) cover most of what gets asked.

What does the Superday actually test?

Three to five interviews covering coding, behavioral depth, and for some teams an open-ended design conversation. Interviewers weight how you reason and communicate tradeoffs heavily, not just whether you reach the optimal solution.