OpenAI technical screen: coding interview
Round 2 of 7 · 13 questions reported
- Length
- 60 min
- Format
- One problem in three to five growing parts filling the hour, usually in CoderPad; ML roles get a Colab notebook with autocomplete off
- They grade
- Working, tested code and sound handling of edge cases; the first two or three parts done cleanly is the usual bar
- AI use
- No AI assistants or autocomplete; documentation is fine
- Heads-up
- Problems are framed as real work: simulations, a memory allocator, an IP range iterator, a credits ledger. A finished, tested solution beats an unfinished clever one. ML and research roles often get ML coding instead (find the bugs in a small transformer, vectorize a NumPy loop), and some infra loops use a 75-minute systems coding round on concurrency.
Questions OpenAI asks in this round 13
- Threshold infection spread on a gridSep 2026 · very often
- Find noisy annotators and retrain a classifier without themSep 2026 · often
- Memory allocator with malloc, free and coalescingAug 2026 · often
- Toy language type inference with generics and tuplesJul 2026 · often
- Three-card showdown between two playersSep 2026 · sometimes
- Entropy from logits, stable and then streamingJul 2026 · sometimes
- Shard rebalancing with a per-key replica limitJun 2026 · sometimes
- Restart strategy that cuts a solver's expected run timeMay 2026 · sometimes
- Resolve package versions from dependency constraintsMay 2026 · sometimes
- Engineering management fundamentals: one-on-ones, hiring and performanceAug 2026 · once
- Cross-entropy loss with masking and label smoothing in NumPyJul 2026 · once
- Max-score grid descent with limited two-row jumpsJul 2026 · once
- Design an AI prompt playgroundOct 2025 · once
About the OpenAI technical screen: coding round
What does OpenAI ask in the technical screen: coding round?
The most asked: Threshold infection spread on a grid, Find noisy annotators and retrain a classifier without them and Memory allocator with malloc, free and coalescing. 13 questions reported in all.
How long is the OpenAI technical screen: coding round?
About 60 minutes.
What is the OpenAI technical screen: coding round like?
One problem in three to five growing parts filling the hour, usually in CoderPad; ML roles get a Colab notebook with autocomplete off. They grade: Working, tested code and sound handling of edge cases; the first two or three parts done cleanly is the usual bar.
Can you use AI in the OpenAI technical screen: coding round?
No AI assistants or autocomplete; documentation is fine.