Just finished a Lyft SWE loop and did the online assessment plus two live coding rounds. Here's the format as of early 2026:
Online assessment (OA): CodeSignal-style, 70 minutes, 4 problems. The first two are closer to LeetCode easy-medium: string manipulation, array traversal, nothing exotic. Problems 3 and 4 step up: one graph problem (BFS/DFS, think adjacency list), one that involved some dynamic programming logic around an interval-scheduling-style constraint. No system design in the OA. Time pressure is the main challenge, not raw difficulty.
Scoring cutoff to advance to phone screen is not published but I'd estimate you need to get through at least 3 cleanly. I finished all 4 but the last one had a suboptimal solution (got it working, not optimal time complexity). Still moved forward.
Live coding rounds (2 of them, 45 min each): These are more conversational than the OA. Both my interviewers asked a medium-ish problem and then added a follow-up constraint partway through. One problem involved a graph traversal on a map grid (think: finding shortest path with some cells blocked and some with varying cost). The follow-up was about memory constraints.
Second round was more algorithmic: a sliding window problem that escalated into a variant where the window had irregular bounds based on some condition. Not LeetCode hard exactly but required thinking clearly about invariants.
They use a shared editor (CoderPad or similar). You can use Python, Java, Go, C++. No external libraries.
One thing I noticed: both interviewers wanted me to walk through test cases before running. Don't just type and hit run. They're evaluating how you think, not just whether the code passes.
Overall difficulty: medium. Not brutally algorithmic like some big-tech shops. If you're solid on graphs, trees, sliding windows, and basic DP you're in good shape. They're not asking segment trees or heavy combinatorics.