Went through GitHub's coding rounds as part of a mid-level SWE loop earlier this year, here's the honest breakdown.
There's no traditional OA in the sense of a timed HackerRank link they fire off before talking to you. At least not for experienced hires. The process started with a recruiter call, then a technical phone screen with a live coding component via a shared editor (they use their own tool, obviously), then two coding rounds during the onsite.
Format and difficulty:
The phone screen coding problem was medium difficulty. Graph traversal, BFS. Manageable in 45 minutes if you can just talk while you code. I babbled too much at first and almost ran out of time, so stay aware of the clock.
The onsite coding rounds were also medium, maybe medium-hard. One problem involved string manipulation that was trickier than it looked because of edge cases around Unicode (which, again, very GitHub-relevant given internationalization of repo content). The other was more straightforward trees/recursion.
I didn't see any Leetcode Hard problems. A few people on here mentioned getting dynamic programming questions but I didn't. Could vary by team or interviewer.
Language choice: they let you pick. I used Python, no issues. They care about code correctness and clarity more than algorithmic heroics.
Time management: both rounds were 45-50 minutes. They usually do a 5-10 minute intro, then problem, then time at the end for questions. So you have maybe 30-35 minutes of actual coding time. Plan accordingly.
One thing I appreciated: both interviewers were clear about whether they wanted an optimal solution or a working solution first. That's not universal across companies. Made it easier to pace myself.
Overall: if you can solve Leetcode mediums comfortably without panicking, you'll be fine on the coding rounds. The behavioral and system design are where I'd focus more prep energy for the senior track.