OpenAI Interview Question Bank
OpenAI pairs an unusually high raw-coding bar with explicit no-AI-tools rules during most interviews — a deliberate contrast with the product. Phone screens often run a single multi-part simulation for the full hour, and senior loops add deep ML and research-grounded system-design rounds that probe mechanism, not vocabulary.
225
questions
Last updated
Before you start
Showing 53 of 53
Implement an At-Least-Once Task Scheduling Queue with DLQDesign a Spreadsheet with Formula Evaluation and Cycle DetectionExtensible Card Hand Comparison Game1-Nearest Neighbor Classification with Manhattan DistanceGPU Credits with ExpirationEvaluate Hierarchical Alert Routing RulesCompare Three-Card Poker Hands with Optional Flush RuleGPU Credit Ledger with Out-of-Order EventsPrefix-Product AutogradIn-Memory Key-Value Store with Write-Ahead Log RecoveryEvent Counting in a 15-Minute Sliding WindowDesign a Spreadsheet with Formula Dependencies and Cycle DetectionImplement Causal Self-Attention with a KV CacheGuess a Secret Number with One-Call Delayed FeedbackImplement Cross-Entropy Loss1-Nearest Neighbor ClassificationPrefix Product AutogradDependency-Aware Multi-Agent Request SchedulerGPU Credit System with Validity WindowsIn-Memory Key-Value Store with SerializationVectorized 1-Nearest Neighbor and Neural-Network FormulationImplement a Work Queue with Leases, Retries, and a Dead-Letter QueueVersioned User Follow Graph with SnapshotsSnapshot SetTopological Sort of Task DependenciesImplement a Durable Key-Value Store with File SegmentationIn-Memory KV Store with Log-Based RecoveryImplement a Memory Allocator with First-Fit and Best-FitFriend Circles / Number of ProvincesCount Machines in a Cluster Tree and Recover Tree Topology