OpenAI hero backdrop - San Francisco skyline
LiveRefreshed weekly

The OpenAI interview cheat code nobody talks about

77 verified questions from the last 14 days, the same ones OpenAI is asking candidates this week. Most applicants never see them.

See What They're Asking

New questions are added every 7 days.

77
Fresh questions
3.2K+
Engineers using this
$325K
Median offer at stake
Scroll
1455 3rd Street · San Francisco
The Process

Six stages, and the coding rounds look nothing like LeetCode

01
30 minVideo call

Recruiter screen

Short, and mostly about whether you want this specific job. It also routes you to a track, and Applied, Research, Infrastructure and Safety run materially different loops.

  • Have an opinion about the products.

    Saying you find AI exciting is the answer everyone gives. Naming a thing you built on the API, and what broke, is not.

  • Ask which team you are being routed to.

    Applied, Research and Infrastructure run different loops. The one you are routed to decides what the onsite tests.

  • Level is decided later than you think.

    OpenAI sets the band after the loop, from the interview feedback, so an early number anchors you with no evidence behind it.

02
20 to 30 minVideo call

Hiring manager screen

A resume deep dive and role fit conversation with the manager, and research candidates walk through their publications. It is a real gate, not a formality.

  • Pick the project you want pulled on.

    The manager will choose a line off your CV and follow it down. Decide in advance which project you want that to be.

  • Know which track you are aiming at.

    Applied, Research, Infrastructure and Safety run materially different loops, and track fit is decided around here.

  • Do not argue seniority yet.

    OpenAI sets the band after the loop, so pushing on level at this stage spends credibility for nothing.

03
60 minYour own editor

Technical screen

One practical problem in a working repo, not a puzzle. You keep your editor, your docs and your search. What is scored is how fast you get something correct running.

  • Get something running early.

    A rough working version at twenty minutes beats an elegant half finished one at sixty. They are watching your loop, not your first draft.

  • You are allowed to look things up.

    Docs and search are open in most rounds. Coding assistants are not, and the candidate instructions say which is which.

  • Write the check before you tune it.

    Most candidates optimise code they have never run. Five lines of verification first is the cheapest score of the day.

04
Varies by teamTake home or pairing

Skills assessment

Format varies by team and the variability is itself the finding: a platform assessment, a paid 48 hour take home worth around $1,000, or a pair coding session.

  • Treat a take home like a pull request.

    It is reviewed for shipping speed, test coverage, documentation and evaluation discipline, not simply whether the thing runs.

  • Ask which format you will get.

    A platform test, a 48 hour take home and a live pairing session need different preparation, and recruiters will tell you which one is coming.

  • Budget the 48 hours deliberately.

    The take home is paid and time boxed. Scope it so the last hours go to tests and a README rather than one more feature.

05
4 to 6 roundsOne to two daysVirtual

Onsite loop

Applied runs two coding rounds, one of them debugging or extending an existing codebase, plus ChatGPT adjacent system design. Infrastructure goes to GPU scheduling and inference at scale, research to paper critique and maths.

  • Bring a project with real numbers.

    Latency, cost per request, GPU hours, failure rate. The deep dive interviewer asks for one inside the first ten minutes.

  • Know how training and serving differ.

    Even product roles get asked how a model is served, batched and rate limited. Nobody expects you to derive attention from scratch.

  • The ambiguity is deliberate.

    The problem statement is thin on purpose. Asking what should happen at the edges is scored, and guessing quietly is not.

06
45 to 60 minWith a leaderScored

Mission round and offer

A scored conversation with the hiring manager and often leadership about why this work and where you would stop, then about a week of committee review and team matching before an offer.

  • Read what they have published.

    The model spec and the preparedness framework are the reference points people quote back, and it shows when you have read neither.

  • Have a line you would not cross.

    Unqualified enthusiasm scores badly and so does doom. They want somebody who can name a tradeoff and then still ship.

  • Ask how the equity actually works.

    Legacy profit participation units converted to ordinary shares in October 2025 and offers made from around January 2026 are conventional RSUs. Ask about vesting and about when you could actually sell.

Levels and Pay

Most of the offer is stock, and the stock is still private

Pick a level to see how the package splits and what the loop asks at that band. The three US tabs sit at parity by policy: one OpenAI requisition lists San Francisco, New York and Seattle at a single band.

LevelExperienceTotal comp
Software EngineerL2Entry0 to 2 yrs~$250K
Software EngineerL3Mid2 to 4 yrs~$325K
Senior EngineerL4Senior4 to 8 yrs~$715K
Staff EngineerL5Staff8 to 12 yrs~$860K
Principal EngineerL6Principal12+ yrs~$1.27M
Live Question Feed

What OpenAI is asking right now

77 questions · new ones added every 7 days
77 Verified
PRACTICAL

Here is a training script that does not converge. Find the bugs, fix them, and say what you would add to catch this earlier.

Technical screen
60 min
Your own editor
Reported 11 times
Open
76 more questions available

The exact questions OpenAI is asking this week, tagged by round and difficulty.

Unlock All 77 Questions

Each question stays in the bank for 14 days, so everything here was asked in the last 14 days. Candidates who see them first pass at 3.2 times the normal rate.

Get Full Access
The Unfair Advantage

Why this actually works

01

Real questions, not guesses

Collected from engineers who sat the loop, then cross validated against other intel sources.

02

Fresh every week

We add new questions every 7 days and retire each one after 14 days, so the bank turns over twice in the window it covers.

03

Covers every round

Practical coding, system design, the skills assessment, the onsite loop and the mission round.

04

120+ companies tracked

Also interviewing at Google, Meta or Microsoft? We have their questions too.

05

Difficulty and pattern data

Tagged by difficulty, round and category, so you know where prep time goes.

06

Candidate verified only

Reported by a candidate, then cross verified against multiple data points.

3.2×
Higher pass rate
120+
Companies tracked
3.2K+
Engineers using this
FAQ

Common questions

Directly from engineers who interviewed at OpenAI, then cross validated against other intel sources into a stack ranked list.

We add new questions every 7 days and retire each one after 14 days, so the bank turns over twice in the window it covers.

Rarely. The screens are production style problems in a working repo, and you keep your editor and your documentation. Pattern grinding is close to useless here.

In live rounds, no. Search and documentation are open, coding assistants are not, and the candidate instructions state the difference.

A scored 45 to 60 minute conversation with the hiring manager and often leadership about deploying AGI safely. Generic enthusiasm is a common rejection reason; they push back and see whether your view holds.

Around $325K in San Francisco, or about £179K in London. About a quarter of that is private stock, which only becomes cash at a tender offer or a listing.

Four to eight weeks, and longer for research and senior roles. Most of the waiting is committee review and team matching, which happen after the hiring decision.

No. GothamLoop covers 120+ companies including Google, Meta, Amazon, Microsoft and Apple.

Pricing plan

Get used to winning

Monthly Access
Month by month access to the most accurate SWE interview question list on the planet
$199
Per month
Buy now
3h : 45m : 23s
Lifetime Access
Unlimited access to all current and future SWE interview questions, forever.
Save 60%
$800
$329
One time
Buy now
Company Interview Guides

Every company we track, each with its own loop, levels and live question feed.

AmazonAndurilAnthropicDatabricksGoogleJane StreetJ.P. MorganJump TradingMetaMicrosoftNetflixOpenAIPalantirSalesforceStripeTikTokUberWaymo
$472k
Palo Alto, CA
$748k
Mountain View, CA
58%
Increase
$382k
San Francisco, CA
$561k
San Francisco, CA
47%
Increase
$347k
New York, NY
$463k
New York, NY
33%
Increase
$426k
San Jose, CA
$618k
San Francisco, CA
45%
Increase
$472k
Palo Alto, CA
$748k
Mountain View, CA
58%
Increase
$382k
San Francisco, CA
$561k
San Francisco, CA
47%
Increase
$347k
New York, NY
$463k
New York, NY
33%
Increase
$426k
San Jose, CA
$618k
San Francisco, CA
45%
Increase
How it worksPricingCompaniesSign inSubscribe
© 2026 Gotham Loop. All rights reserved.
Terms of Service