Did the VMware SWE loop twice, once in 2024 and once in early 2026 for two different orgs (first was Cloud Foundation, second was vSphere compute). Format changed a bit between the two. Here's what the coding side looks like now.
Online Assessment (pre-loop): HackerRank platform, 90 minutes, 2 problems. First was medium-difficulty, string or array manipulation. Second was medium-hard, usually graph or dynamic programming. The 2026 version I did had a sliding window problem and a shortest-path variant with a twist. Nothing I'd call "hard" on LeetCode scale. More like "consistent medium with one edge case they specifically want to see you handle."
No proctoring beyond browser lockdown. You have access to your language's standard library.
Coding rounds in the onsite (phone + final): Two coding rounds during the full loop. Each 45-50 minutes. One interviewer per round. Problems were again in the medium range. I had: A tree problem (level order traversal variant with some constraints) A design-adjacent coding problem where you implement a simplified in-memory LRU cache
They care a lot about clean code, not just correctness. I was asked to talk through my complexity analysis and they pushed back when I gave a sloppy O(n) answer without qualifying which n. Space complexity matters too.
Language: I used Python and Go in different rounds. No preference expressed. Java and C++ also commonly used.
Compare to FAANG: Easier than Google/Meta. Closer to Amazon SDE2 level. Don't expect LeetCode hard. Do expect clean, readable code and good communication throughout.
If you're aiming for senior (L5 equivalent post-Broadcom), two solid mediums well-explained should clear the bar. Sloppy mediums probably won't.