Amazon Online Assessment (OA): what the 2026 test really asks
For most candidates the Amazon Online Assessment is the first real gate of the loop, and it is unlike the rounds that follow it: there is no interviewer, no voice, no follow-up — just a clock, two problems, and a grader that runs your code against test cases you can see and test cases you cannot. That changes what you optimise for. An onsite round rewards how you talk through a solution; the Amazon OA rewards correct, complete code delivered on time, because the only thing being read is whether your submission passes the hidden tests before the ninety minutes run out.
This is a map of that assessment for 2026 — how the Amazon OA questions are structured, the handful of patterns they keep reaching for, the pacing discipline that decides a pass, and where the OA sits before the phone screen and onsite. It is framed the way the test is framed: not "here are the exact questions" but "here is the pattern, here is the kind of problem it wears, and here is the edge case a hidden test will punish if you miss it." Treat it as a prep plan, because the OA is won on finishing — two problems mostly right beats one perfect and one blank.
How the Amazon OA runs (2026)
The Amazon OA is a timed, auto-graded coding test you take remotely, early in the process — usually before any human speaks to you. In the modelled version of this round you get 90 minutes and two problems, each graded against visible and hidden test cases, so correctness is checked exactly the way the real assessment checks it: your code has to actually pass, not merely look right. The two problems sit mostly in the medium band — a graph traversal, a dynamic-programming count, a matrix or array problem — and the quiet signal the grader reads is whether you produce complete, correct solutions under time pressure.
The OA has no interviewer to nudge you, which is exactly what makes it hard. "Your solution passes the three visible tests and fails two hidden ones" is the message that ends more OAs than a wrong approach does — because the hidden tests are the empty input, the single element, the duplicate, the cycle, the integer that overflows. The assessment is really testing whether you handle the cases nobody showed you.
The rounds on Zynter are modelled on how the Amazon loop runs, not scraped from any candidate's recall. What matters for prep is the shape: two problems, a clock, real code that compiles and runs against visible and hidden tests, and a debrief afterwards that shows you where the time and the test cases went.
What the OA is really testing
Because it is auto-graded, the OA leans on three of the four debrief dimensions harder than the onsite does. Problem solving — do you reach a correct, optimal-enough approach without a hint? Technical execution — does your code actually run and pass the hidden tests, including the edge cases nobody displayed? Time management — do you budget ninety minutes across two problems so you finish, rather than sink all of it into one? The fourth, communication, barely registers here — it is the onsite's job — which is why the OA can feel deceptively mechanical and then fail you on a case you never saw. Reading this guide teaches the patterns; it does not teach you to pace two problems under a live clock.
The question types on the Amazon OA
You do not need five hundred problems; you need the handful of coding interview patterns the OA keeps reusing, and the habit of spotting which one a prompt is wearing. Amazon leans on graphs, dynamic programming, and array/string manipulation at medium difficulty. The table is the shortlist: the pattern, the kind of problem it shows up in, and the hidden-test trap that decides whether your submission actually passes.
| Pattern | Example problem type | The hidden-test trap |
|---|---|---|
| Arrays & strings (two pointers, sliding window) | Longest substring without repeats, partition equal subset sum | Empty string, all-same characters, the window never shrinking |
| Hashing | Two-sum, group anagrams, first unique character | Duplicate keys, the single-element case |
| Graphs (BFS / DFS, topological sort) | Course schedule, number of islands, pacific atlantic | A cycle, a disconnected node, an empty graph |
| Trees (BFS / DFS) | Level-order, right side view, path sums | A skewed tree's stack depth, a single-node tree |
| Dynamic programming | Unique paths, coin change, word break | The zero/empty base case, an unreachable target |
| Matrix | Valid sudoku, unique paths, set matrix zeroes | A 1×N or N×1 grid, mutating the input in place |
| Heaps / top-K | K closest points, merge k sorted lists | K larger than the input, ties at the boundary |
The two coding problems
Both problems auto-grade the same way, so your job is a correct, complete solution with every edge case covered — the grader does not award partial credit for an elegant idea that crashes on an empty input. Two templates cover a large share of OA problems. The sliding window handles "longest / shortest / count of" over an array or string:
# Sliding window: grow on the right, shrink on the left when the window breaks a rule
left = 0
for right in range(len(s)):
add(s[right])
while invalid(): # a constraint broke — pull the left edge in
remove(s[left]); left += 1
best = max(best, right - left + 1)
Breadth-first search handles level-order traversal and shortest-path-in-an-unweighted-graph, and a close DFS variant handles island counting and course-schedule cycle checks:
from collections import deque
def bfs(start, neighbors):
seen, q = {start}, deque([start])
while q:
node = q.popleft()
for nxt in neighbors(node): # explore one frontier at a time
if nxt not in seen:
seen.add(nxt); q.append(nxt)
return seen
Have both automatic, because on a 90-minute clock the time you save not re-deriving a template is the time you spend covering the hidden-test edge cases.
Pacing the 90 minutes
Pacing is the skill most candidates underestimate, and it is the one the OA punishes hardest. Two problems, ninety minutes, auto-graded — the maths is simple and unforgiving: a single problem you cannot finish is worse than two you solve at eighty percent, because a blank submission scores nothing while a nearly-complete one often passes most tests.
- Read both problems first (2–3 minutes). Start with the one you see the approach for; a fast first solve banks time and confidence.
- Get something running early. A brute-force that passes the visible tests is worth more than an optimal solution still in your head at the 80-minute mark — then optimise if time allows.
- Budget the clock. Roughly forty minutes a problem, ten held back. If you are 45 minutes in and stuck on the first, move: a partial second solve beats a perfect first and a blank.
- Spend the last ten minutes on edge cases, not polish. Empty input, one element, duplicates, a cycle, overflow — these are the hidden tests. Walk each solution through them before the clock dies.
The move that reads as prepared: treat the edge cases as part of the solution, not an afterthought, because on an auto-graded test the case you did not handle is the case that fails you silently.
Coding
Amazon · Online Assessment · 90 min
Run the Amazon OA timed, two problems against hidden tests
Beyond the OA: the phone screen and onsite
The OA is a gate, not the whole loop. Clear it and the process becomes a phone screen and an onsite of several back-to-back rounds — coding, system design, and Leadership Principles — where a human is in the room (by voice) and the grade shifts. There, how you talk through a solution matters as much as whether it runs: you narrate the plan, handle edge cases out loud, and answer the follow-up after every submission. The same problem that is a clean auto-graded pass can fail an onsite if you cannot explain why your approach is correct and what it costs. Prepare the OA for correctness and pacing; prepare the onsite for defending your answer out loud.
How to prepare: a two-week plan
You do not need a hundred new problems in the last fortnight. You need the patterns automatic, the edge cases reflexive, and the pacing rehearsed. Here is a realistic two-week pass for Amazon OA prep.
- Days 1–3: Drill arrays, strings and hashing — two pointers, sliding window, the hash-map lookup — until you recognise them on sight and can code them without re-deriving the template.
- Days 4–8: Live in graphs and trees — BFS and DFS, topological sort, island counting, path sums — and dynamic programming. These anchor most OA problem sets.
- Days 9–11: Edge cases and pacing. For every solution, list the empty/single/duplicate/cycle/overflow cases before you call it done, and practise a strict per-problem time budget.
- Days 12–14: Full timed OAs — two problems, 90-minute clock, auto-graded against visible and hidden tests — then review where the time and the failed cases went. The goal is to finish both, correct, on pace.
The most common mistake in the last week is grinding more problems without the clock. By day 12, the bottleneck is not knowing another algorithm — it is finishing two correct solutions in ninety minutes. If you want raw volume to drill the problems themselves, PipeCode is a sister platform built for exactly that; use it for reps, then come back here to rehearse the timed, graded format.
Coding
Amazon · Onsite coding · 60 min
After the OA, practise an onsite coding round with follow-ups
The mistakes that fail an Amazon OA
The OA fails more candidates on process than on algorithms. Master these five habits and a passing approach actually becomes a passing submission.
- Submitting without the edge cases. The hidden tests are the empty input, the single element, the duplicate, the cycle, the overflow. Handle them before you submit, not after you fail.
- Sinking 90 minutes into one problem. Budget the clock; a partial second solve beats a perfect first and a blank. Move when you are stuck.
- Optimising before it runs. Get a correct brute force passing the visible tests first, then improve it. An optimal idea that doesn't compile scores zero.
- Ignoring complexity. A correct but quadratic solution can time out on the large hidden inputs. Know your time and space cost and whether it survives the constraints.
- Not reading both problems first. Starting the harder one blind burns the clock; scan both and open with the one you see.
Each of these maps to the OA's three live dimensions — problem solving, technical execution and time management. Reading the list teaches the habits; practising them under a real 90-minute clock is what makes them automatic.
Frequently asked questions
What is the Amazon Online Assessment?
The Amazon Online Assessment (OA) is a timed, auto-graded coding test taken remotely, usually as the first step of the loop before any interviewer speaks to you. In the modelled version of the round it is 90 minutes with two coding problems, each checked against visible test cases you can run and hidden test cases revealed only on submission. Because there is no interviewer, it rewards correct, complete code delivered on time rather than how you talk through a solution. The problems sit mostly in the medium band — graphs, dynamic programming, arrays and strings — and the real test is whether you handle the edge cases the hidden tests probe.
How long is the Amazon OA and how many questions?
In the modelled round, the Amazon OA runs 90 minutes with two coding problems, auto-graded against visible and hidden test cases. Pacing is the skill candidates most underestimate: budgeting roughly forty minutes per problem with ten held back for edge cases beats sinking all ninety into one. Getting one problem fully correct and the second most of the way is a better outcome than one perfect answer and a blank, because the grader rewards passing tests, not elegant ideas left unfinished.
What kinds of questions are on the Amazon OA?
Mostly medium-difficulty problems across a handful of patterns: arrays and strings (two pointers, sliding window), hashing, graphs (BFS, DFS, topological sort), trees, dynamic programming, matrix problems and occasionally heaps. Amazon leans harder on graphs and dynamic programming than most. The decisive detail is the hidden tests — empty inputs, single elements, duplicates, cycles and integer overflow — so a pattern you can code but whose edge cases you skip will still fail. Prepare the patterns for recognition and the edge cases for completeness.
How do I pass the Amazon OA?
Finish. Two problems mostly correct beats one perfect and one blank, because a blank submission scores nothing. Read both problems first and start with the one whose approach you see; get a correct brute force running against the visible tests before optimising; budget the clock so you never sink all ninety minutes into one problem; and spend the last ten minutes walking each solution through the empty, single-element, duplicate, cycle and overflow cases, because those are the hidden tests. Know the time and space complexity of each solution so it does not time out on the large hidden inputs.
Does the Amazon OA have an interviewer or follow-up questions?
No — the OA is auto-graded with no interviewer, which is what distinguishes it from the rounds that follow. There is no one to nudge you when you are stuck and no chance to explain your reasoning; only your submitted code is read, and only against the test cases. The voice follow-ups — "why that data structure?", "what breaks at a hundred million rows?" — belong to the phone screen and onsite rounds that come after the OA. That is why OA prep is about correctness and pacing, while onsite prep is about defending your answer out loud.
How do I practise the Amazon OA out loud?
The OA itself is silent, but the onsite rounds after it are not, and practising both is the point. On Zynter, the Coding track runs live rounds modelled on the Amazon loop — including a 90-minute Online Assessment with two problems auto-graded against visible and hidden tests, and 60-minute onsite coding rounds where a voice interviewer asks the follow-ups, then a debrief on problem solving, communication, technical execution and time management. Practise the OA for pacing and correctness, then the onsite for defending it out loud. The score is a practice signal, not a hiring decision.
Practise it on Zynter
You can solve the problems. The OA is won on finishing two of them, correct, before the clock dies.
Pick the Amazon Online Assessment round, write code that runs against visible and hidden test cases on a 90-minute clock, and see where the time and the failed cases went in the debrief — then practise the onsite rounds out loud. Zynter.ai is on-demand mock interviews modelled on real company loops, with 150+ rounds across 20 companies including the Amazon Online Assessment and onsite coding rounds (Sep 2026).
Top comments (0)