The Goldman Sachs Software Engineer Interview Starts With a HackerRank OA: What to Expect

Two problems. Somewhere between sixty and ninety minutes. That is the entire gate between your resume and the first human being at this bank who will ever read your name out loud.

I sent 419 applications the year I was trying to claw out of a contract role, and four of them turned into anything at all. The Goldman Sachs software engineer interview was one of those four, and it did not start with a recruiter call or a warm intro or someone from my school who worked there — because nobody from my school worked there. It started with an automated HackerRank link that showed up in my inbox on a Tuesday morning. No conversation. No context. A link and a clock.

If you are reading this because your link already arrived, this piece is about that link specifically. Not the Superday. Not the offer. The online assessment is where most applicants to Goldman's Engineering division quietly stop existing in the process, and almost none of them ever find out why.

What you're actually applying to, before you open anything

Engineering at Goldman is a division, not a team. That matters more than it sounds. It means the OA is a shared front door — the same assessment funnels people toward wildly different desks, and the culture, tooling, and day-to-day work on the other side varies by where you land. You are not testing into one job. You are testing into a pool.

The realistic-target math is genuinely favorable if you didn't come up through the FAANG pipeline. Goldman sponsors H1B. It hires into New York, Dallas, and Salt Lake City. And the loop's difficulty sits at Medium — not because the bank is soft, but because the algorithmic ceiling is lower than what Meta or a quant shop will throw at you. Goldman is not screening for the person who invents a novel solution under pressure. It is screening for the person who reliably ships a correct one, three times in a row, without falling apart.

That distinction shapes everything about the OA.

Stage one: the link arrives

The invite is automated and it is thin. It will tell you the platform (HackerRank) and give you a window to complete it. It will not tell you which team you're being considered for, who reviewed your resume, or what a passing score looks like. I don't have a documented retake policy for Goldman, and I'm not going to invent one — assume you get one attempt and treat it that way.

What you control at this stage is when you sit down. The window is usually a few days wide. People burn this advantage constantly: link arrives at 6:40am, adrenaline spikes, they open it at 6:45am on a laptop at the kitchen table with a roommate making coffee. Don't. Take the two or three days. Sit down at the hour your brain actually works.

What's on the assessment

Two problems. Roughly 60–90 minutes total. The difficulty band reported by readers is Easy-to-Medium, with an occasional Medium-Hard dynamic programming problem showing up as one of the two.

That shape is the whole strategy. Two problems in ninety minutes is not a sprint — it is thirty to forty minutes each with slack. But the variance between an Easy pairing and an Easy-plus-Medium-Hard-DP pairing is enormous. If both problems are approachable, the assessment is a formality. If one of them is the DP, that problem is the assessment, and the other one is a tax you need to pay in under twenty minutes so you have time for the real thing.

The patterns that actually show up

Here are three specific things readers reported in 2026, which is more than you will get from a single Glassdoor glance:

A two-window DP on a price series. The reported shape: find the maximum profit from at most two non-overlapping trade windows in an array of prices. If you have seen the classic buy-and-sell-stock family, you know this one — it's the two-transaction variant, and it is the single most on-brand problem a bank could possibly ask. The naive approach is to split the array at every index and run single-transaction on both halves, which is O(n²) and will time out on the hidden tests. The O(n) version tracks four rolling states:

def max_profit_two(prices):
    buy1 = buy2 = float('-inf')
    sell1 = sell2 = 0
    for p in prices:
        buy1  = max(buy1, -p)          # bought once
        sell1 = max(sell1, buy1 + p)   # sold once
        buy2  = max(buy2, sell1 - p)   # bought twice
        sell2 = max(sell2, buy2 + p)   # sold twice
    return sell2

Four lines of state you should be able to reconstruct from memory, not rederive under a timer. If you internalize one thing before your assessment, make it this family — one transaction, two transactions, k transactions, with cooldown, with fee.

First non-repeating order ID in a stream. Reported often. This is a hash-map-plus-ordering problem dressed in finance clothing: counts in a dictionary, insertion order preserved, return the first ID with a count of one. Trivial if you've seen it. A mess if you try to solve it with nested loops.

And the one that isn't algorithmic at all: every Superday includes some version of why Goldman, and why engineering here instead of at a tech company? That question doesn't appear on the OA, but knowing it's coming should change how you think about the OA — it is the first move in a process that ends with you defending a choice, not a solution.

Where people actually get cut

Not on brilliance. Here is what killed the candidates I've talked to:

• Hidden test cases on empty and single-element input. Your solution handles the happy path, passes the two visible samples, and fails four hidden ones. Write the three-line guard clause first, before the algorithm.
• Time allocation. Forty-five minutes sunk into problem one, fifteen minutes left for a DP. Set a hard mental cap. A working brute force on both problems beats an elegant solution on one and a blank editor on the other, because HackerRank scores on test cases passed, not on beauty.
• Not reading the I/O format. Some problems hand you parsed arguments; some hand you stdin. Losing eight minutes to a parsing error you introduced is a genuinely stupid way to lose a callback, and it happens constantly.
• Practicing on the wrong surface. If your entire prep has been in an IDE with autocomplete and a debugger, the HackerRank editor will feel like typing with gloves on. Do at least three full-length timed sessions in the actual browser environment.

What happens if you pass

The OA is followed by a recruiter check — about twenty minutes, mostly logistics and team routing, since Engineering is a division and the desks differ. Then the Superday: three interviews in one block, roughly three hours, covering coding, a systems or practical round, and behavioral.

That structure is the tell. Goldman's process rewards preparation and stamina more than it rewards raw insight. The people who fail Superday usually solve the first problem fine and are cognitively spent by interview three. Practice interviewing three times in a row, not solving problems three times in a row. Those are different skills.

The honest comp read

Base range across the posted roles runs $115K–$180K, as of this writing. Notice the empty equity column — the data here carries no RSU figures, and you should assume your package is base plus a bonus that is discretionary and moves with the firm's year, not a predictable grant that vests on a schedule. That is a real structural difference from a tech offer, and it is the thing most candidates coming from a tech mindset misread.

The honest framing: the brand is worth more than the number. Two years at Goldman re-brands a non-target resume in a way that almost nothing else at this difficulty level does. The trade is comp that lags tech and an engineering culture that varies by desk. Bonus and level are the negotiable pieces; base tends to sit inside a band.

Verdict

Target this if you're outside the pipeline and you need a name on your resume that opens doors you can't currently knock on — and if a Medium algorithmic bar plus a stamina test is a fight you can win. Skip it if you're optimizing purely for total compensation, or if you want equity upside, because the numbers above are the numbers.

The highest-leverage prep, if you have one week: the buy-and-sell-stock DP family, until you can write the four-state version cold.

The OA is the only part of this process that doesn't care where you went to school — which makes it the cheapest door anyone in tech will ever hand you.